Update flake and fix it

Just for validation that delete does the right thing

.envrc

@@ -0,0 +1,2 @@
+use nix
+

.woodpecker/build.yml

@@ -5,28 +5,19 @@ platform: linux/amd64
 
 pipeline:
   unitTests:
-    image: "docker.bics.ga/rei_ci/rust-sccache:latest-amd64"
+    image: "rust:1.65.0"
     pull: true
     commands:
       - DEBIAN_FRONTEND=noninteractive apt-get -qq update > /dev/null
       - DEBIAN_FRONTEND=noninteractive apt-get -yqq install pkg-config libssl-dev build-essential libsqlite3-dev > /dev/null
       - cargo build --all
       - cargo test --all
-      - sccache --show-stats
-    environment:
-      RUSTC_WRAPPER: /usr/local/bin/sccache
-      SCCACHE_S3_USE_SSL: "true"
-      SCCACHE_ENDPOINT: "richie.m4.tanukitsu.net:443"
-    secrets:
-      - sccache_bucket
-      - aws_access_key_id
-      - aws_secret_access_key
     when:
       event: [push, pull_request]
 
 
   testSuite:
-    image: "docker.bics.ga/rei_ci/rust-sccache:latest-amd64"
+    image: "rust:1.65.0"
     commands:
       - DEBIAN_FRONTEND=noninteractive apt-get -qq update > /dev/null
       - DEBIAN_FRONTEND=noninteractive apt-get -yqq -o=Dpkg::Use-Pty=0 install pkg-config libssl-dev build-essential libsqlite3-dev python3.9 python3.9-venv postgresql postgresql-client mariadb-server mariadb-client zstd lz4 > /dev/null
@@ -41,20 +32,11 @@ pipeline:
       - python3.9 -m venv testsuite/.venv
       - ./testsuite/.venv/bin/pip install ./testsuite ./datman-helper-postgres ./datman-helper-mysql
       - cd testsuite && . .venv/bin/activate && TEST_POSTGRES=$(hostname),testsuitedb,root TEST_MYSQL=$(hostname),testsuitemydb,root green
-      - sccache --show-stats
-    environment:
-      RUSTC_WRAPPER: /usr/local/bin/sccache
-      SCCACHE_S3_USE_SSL: "true"
-      SCCACHE_ENDPOINT: "richie.m4.tanukitsu.net:443"
-    secrets:
-      - sccache_bucket
-      - aws_access_key_id
-      - aws_secret_access_key
     when:
       event: [push, pull_request]
 
   deployManual:
-    image: "docker.bics.ga/rei_ci/mdbook:latest-amd64"
+    image: "docker.emunest.net/rei_ci/mdbook:latest-amd64"
     pull: true
     when:
       branch:

Cargo.lock

@@ -404,7 +404,7 @@ dependencies = [
 
 [[package]]
 name = "datman"
-version = "0.6.0-alpha.4"
+version = "0.6.0-alpha.5"
 dependencies = [
  "anyhow",
  "arc-interner",
@@ -1733,7 +1733,7 @@ dependencies = [
 
 [[package]]
 name = "yama"
-version = "0.6.0-alpha.4"
+version = "0.6.0-alpha.5"
 dependencies = [
  "anyhow",
  "blake",

datman-helper-postgres/datman_helper_postgres/backup.py

@@ -39,10 +39,7 @@ def cli():
     # The process (if any) that is our LZ4 decompressor.
     lz4_process = None
 
-    dump_command = [
-        "pg_dump",
-        database_to_use
-    ]
+    dump_command = ["pg_dump", database_to_use]
 
     if host_to_use is not None:
         if use_lz4:
@@ -63,21 +60,19 @@ def cli():
         # (rather than lz4 covering it).
         command = [
             "ssh",
-            f"{user_to_use}@{host_to_use}" if user_to_use is not None else f"{host_to_use}",
+            f"{user_to_use}@{host_to_use}"
+            if user_to_use is not None
+            else f"{host_to_use}",
             "bash",
             "-o",
             "pipefail",
             "-c",
-            shlex.quote(" ".join(dump_command))
+            shlex.quote(" ".join(dump_command)),
         ]
     elif user_to_use is not None:
         current_username = pwd.getpwuid(os.getuid()).pw_name
         if current_username != user_to_use:
-            command = [
-                "sudo",
-                "-u",
-                user_to_use
-            ] + dump_command
+            command = ["sudo", "-u", user_to_use] + dump_command
         else:
             command = dump_command
     else:

datman/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "datman"
-version = "0.6.0-alpha.4"
+version = "0.6.0-alpha.5"
 authors = ["Olivier 'reivilibre' <olivier@librepush.net>"]
 edition = "2021"
 repository = "https://bics.ga/reivilibre/yama"
@@ -30,7 +30,7 @@ humansize = "1.1.1"
 chrono = "0.4.19"
 itertools = "0.10.1"
 hostname = "0.3.1"
-yama = { path = "../yama", version = "0.6.0-alpha.1" }
+yama = { path = "../yama", version = "0.6.0-alpha.5" }
 metrics = "0.17.1"
 bare-metrics-recorder = { version = "0.1.0" }
 comfy-table = "6.0.0-rc.1"

datman/README.md

@@ -8,5 +8,6 @@ Features:
 * (optional) Compression using Zstd and a specifiable dictionary
 * (optional) Encryption
 * Ability to back up to remote machines over SSH
+* Labelling of files in a backup source; different destinations can choose to backup either all or a subset of the labels.
 
 See the documentation for more information.

datman/src/bin/datman.rs

@@ -28,6 +28,7 @@ use bare_metrics_recorder::recording::BareMetricsRecorderCore;
 use chrono::{DateTime, Local, NaiveDate, NaiveDateTime, TimeZone, Utc};
 use datman::commands::backup::{backup_all_sources_to_destination, backup_source_to_destination};
 use datman::commands::ilabel::interactive_labelling_session;
+use datman::commands::prune::{prune_with_retention_policy, RetentionPolicy};
 use datman::commands::{init_descriptor, pushpull};
 use datman::descriptor::{load_descriptor, SourceDescriptor};
 use datman::get_hostname;
@@ -137,6 +138,13 @@ pub enum DatmanCommand {
         pile_name: String,
     },
 
+    /// Applies a retention policy by removing unnecessary backups.
+    /// Does not reclaim space by itself: use
+    ///   `yama check --apply-gc --shallow`
+    /// & `yama compact`
+    /// to do that.
+    Prune { pile_name: String },
+
     #[clap(name = "_pull_responder_offerer")]
     InternalPullResponderOfferer {
         datman_path: PathBuf,
@@ -410,6 +418,24 @@ fn main() -> anyhow::Result<()> {
                 Box::new(pbar),
             )?;
         }
+
+        DatmanCommand::Prune { pile_name } => {
+            let descriptor = load_descriptor(Path::new(".")).unwrap();
+            let retention_policy = descriptor
+                .retention
+                .context("No retention policy set in descriptor")?;
+            let dest_desc = &descriptor.piles[&pile_name];
+
+            let pile_desc = load_pile_descriptor(&dest_desc.path)?;
+
+            prune_with_retention_policy(
+                &dest_desc.path,
+                &pile_desc,
+                &RetentionPolicy::from_config(retention_policy),
+                true,
+            )?;
+        }
+
         DatmanCommand::InternalPullResponderOfferer {
             datman_path,
             pile_name,

datman/src/commands.rs

@@ -20,12 +20,13 @@ use std::fs::File;
 use std::io::Write;
 use std::path::Path;
 
-use crate::descriptor::{Descriptor, SourceDescriptor};
+use crate::descriptor::{Descriptor, RetentionPolicyConfig, SourceDescriptor};
 
 pub mod backup;
 pub mod extract;
 pub mod ibrowse;
 pub mod ilabel;
+pub mod prune;
 pub mod pushpull;
 pub mod report;
 
@@ -51,6 +52,12 @@ pub fn init_descriptor(path: &Path) -> anyhow::Result<()> {
         sources: source,
         piles: Default::default(),
         remote_hosts: Default::default(),
+        retention: Some(RetentionPolicyConfig {
+            daily: 14,
+            weekly: 12,
+            monthly: 24,
+            yearly: 9001,
+        }),
     })?;
 
     datman_toml_file.write_all(&bytes)?;

datman/src/commands/backup.rs

@@ -17,7 +17,9 @@ along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 
 use crate::descriptor::{Descriptor, DestPileDescriptor, SourceDescriptor, VirtualSourceKind};
 use crate::get_hostname;
-use crate::labelling::{label_node, load_labelling_rules, str_to_label, Label, State};
+use crate::labelling::{
+    label_node, load_labelling_rules, str_to_label, Label, LabellingRules, State,
+};
 use crate::tree::{scan, FileTree, FileTree1};
 use anyhow::{anyhow, bail};
 use arc_interner::ArcIntern;
@@ -76,8 +78,8 @@ pub fn open_stdout_backup_process(
 pub fn label_filter_and_convert(
     tree: FileTree1<()>,
     descriptor: &Descriptor,
-    desc_path: &Path,
     source_name: &str,
+    rules: &LabellingRules,
     dest: &DestPileDescriptor,
 ) -> anyhow::Result<Option<TreeNode>> {
     info!("Labelling.");
@@ -87,8 +89,7 @@ pub fn label_filter_and_convert(
         .iter()
         .map(|l| Label(ArcIntern::new(l.clone())))
         .collect();
-    let rules = load_labelling_rules(desc_path, source_name)?;
-    label_node("".to_owned(), None, &mut tree, &labels, &rules)?;
+    label_node("".to_owned(), None, &mut tree, &labels, rules)?;
 
     let included_labels: HashSet<Label> = dest.included_labels.iter().map(str_to_label).collect();
 
@@ -150,16 +151,20 @@ pub fn backup_source_to_destination<PT: ProgressTracker>(
             cross_filesystems,
         } => {
             info!("Looking to backup {} to {}", source_name, dest_name);
+            let rules = load_labelling_rules(desc_path, source_name)?;
+            let exclusions = rules.get_exclusions_set(directory);
+
             info!("Scanning.");
-            let tree = scan(directory, !*cross_filesystems)?
+            let tree = scan(directory, !*cross_filesystems, &exclusions)?
                 .ok_or_else(|| anyhow!("Source does not exist."))?;
 
             let absolute_source_path = desc_path.join(directory);
             let absolute_dest_path = desc_path.join(&dest.path);
             let pile_descriptor = load_pile_descriptor(&absolute_dest_path)?;
             let pile = open_pile(&absolute_dest_path, &pile_descriptor)?;
+
             let root = if let Some(root) =
-                label_filter_and_convert(tree, descriptor, desc_path, source_name, dest)?
+                label_filter_and_convert(tree, descriptor, source_name, &rules, dest)?
             {
                 root
             } else {

datman/src/commands/ibrowse.rs

@@ -15,6 +15,7 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
+use std::collections::BTreeSet;
 use std::path::Path;
 
 use anyhow::{anyhow, bail};
@@ -80,7 +81,7 @@ pub fn session(path: &Path, source_name: String) -> anyhow::Result<()> {
     };
 
     println!("Scanning source; this might take a little while...");
-    let mut dir_scan: FileTree1<Option<State>> = scan(directory, one_filesystem)?
+    let mut dir_scan: FileTree1<Option<State>> = scan(directory, one_filesystem, &BTreeSet::new())?
         .ok_or_else(|| anyhow!("Empty source."))?
         .replace_meta(&None);
 

datman/src/commands/ilabel.rs

@@ -15,6 +15,7 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
+use std::collections::BTreeSet;
 use std::io;
 use std::io::{StdinLock, Stdout, Write};
 use std::path::Path;
@@ -192,7 +193,7 @@ pub fn interactive_labelling_session(path: &Path, source_name: String) -> anyhow
         let my_hostname = get_hostname();
         let mut dir_scan = if &my_hostname == hostname {
             info!("Scanning source; this might take a little while...");
-            scan(directory, !*cross_filesystems)?
+            scan(directory, !*cross_filesystems, &BTreeSet::new())?
                 .ok_or_else(|| anyhow!("Empty source."))?
                 .replace_meta(&None)
         } else {
@@ -212,6 +213,7 @@ pub fn interactive_labelling_session(path: &Path, source_name: String) -> anyhow
                 &mut write,
                 directory.as_ref(),
                 !*cross_filesystems,
+                &BTreeSet::new(),
             )?
             .ok_or_else(|| anyhow!("Remote scan failed (does the directory exist?)"))?
             .replace_meta(&None);

datman/src/commands/prune.rs

@@ -0,0 +1,220 @@
+use crate::commands::backup::split_pointer_name;
+use crate::descriptor::RetentionPolicyConfig;
+use anyhow::{bail, Context};
+use log::info;
+use std::collections::{BTreeMap, BTreeSet};
+use std::io;
+use std::path::Path;
+use yama::commands::open_pile;
+use yama::operations::remove_pointer_safely;
+use yama::pile::PileDescriptor;
+
+pub struct RetentionBand {
+    pub interval_s: u64,
+    pub number_to_retain: u32,
+}
+
+pub struct RetentionPolicy {
+    pub retention_bands: Vec<RetentionBand>,
+}
+
+const DAY: u64 = 86400;
+const WEEK: u64 = 7 * DAY;
+const MONTH: u64 = 31 * DAY;
+const YEAR: u64 = 365 * DAY;
+
+impl RetentionPolicy {
+    pub fn from_config(descriptor: RetentionPolicyConfig) -> RetentionPolicy {
+        let mut policy = RetentionPolicy {
+            retention_bands: vec![],
+        };
+
+        if descriptor.daily != 0 {
+            policy.retention_bands.push(RetentionBand {
+                interval_s: DAY,
+                number_to_retain: descriptor.daily,
+            });
+        }
+
+        if descriptor.weekly != 0 {
+            policy.retention_bands.push(RetentionBand {
+                interval_s: WEEK,
+                number_to_retain: descriptor.weekly,
+            });
+        }
+
+        if descriptor.monthly != 0 {
+            policy.retention_bands.push(RetentionBand {
+                interval_s: MONTH,
+                number_to_retain: descriptor.monthly,
+            });
+        }
+
+        if descriptor.yearly != 0 {
+            policy.retention_bands.push(RetentionBand {
+                interval_s: YEAR,
+                number_to_retain: descriptor.yearly,
+            });
+        }
+
+        policy
+    }
+
+    /// Returns the set of snapshots to remove.
+    pub fn apply_returning_prunable(
+        &self,
+        snapshots_by_unix_time: BTreeMap<u64, String>,
+    ) -> BTreeSet<String> {
+        if snapshots_by_unix_time.is_empty() {
+            return BTreeSet::new();
+        }
+        let mut snapshots_included: BTreeSet<u64> = BTreeSet::new();
+
+        // Always mark the most recent snapshot as retained!
+        let last_snapshot = snapshots_by_unix_time.keys().rev().next().unwrap();
+        snapshots_included.insert(*last_snapshot);
+
+        let now_time = *last_snapshot;
+
+        for band in &self.retention_bands {
+            for multiple in 1..=band.number_to_retain {
+                let target_time = now_time - (multiple as u64) * band.interval_s;
+                if let Some((k, _)) = snapshots_by_unix_time.range(0..=target_time).rev().next() {
+                    snapshots_included.insert(*k);
+                }
+            }
+        }
+
+        // Find all prunable (unincluded) snapshots.
+        snapshots_by_unix_time
+            .into_iter()
+            .filter(|(k, _v)| !snapshots_included.contains(k))
+            .map(|(_k, v)| v)
+            .collect()
+    }
+}
+
+pub fn prune_with_retention_policy(
+    pile_path: &Path,
+    pile_desc: &PileDescriptor,
+    policy: &RetentionPolicy,
+    prompt_first: bool,
+) -> anyhow::Result<()> {
+    let pile = open_pile(&pile_path, &pile_desc).context("Failed to open pile")?;
+
+    let pointers = pile
+        .list_pointers()
+        .context("Failed to list pointers in pile")?;
+
+    let mut pointers_to_keep: BTreeSet<String> = pointers.iter().cloned().collect();
+
+    let pointers_to_remove = get_prunable_pointers(&policy, pointers);
+
+    for remove in &pointers_to_remove {
+        pointers_to_keep.remove(remove);
+    }
+
+    info!("Gory details:\n---\nKeep: {pointers_to_keep:?}\n---\nRemove: {pointers_to_remove:?}");
+    info!(
+        "{} pointers to remove ({} to keep) based on retention policy.",
+        pointers_to_remove.len(),
+        pointers_to_keep.len()
+    );
+
+    if prompt_first {
+        println!("Would you like to proceed? [y/N]: ");
+        let mut buffer = String::new();
+        let stdin = io::stdin(); // We get `Stdin` here.
+        stdin.read_line(&mut buffer)?;
+        if buffer.trim().to_ascii_lowercase() != "y" {
+            bail!("Aborted by user.");
+        }
+    }
+
+    for to_remove in pointers_to_remove {
+        let res = remove_pointer_safely(&pile, &to_remove).context("removing prunable pointers");
+
+        pile.flush()
+            .context("flushing pile after removing pointers")?;
+
+        res?;
+    }
+
+    Ok(())
+}
+
+fn get_prunable_pointers(policy: &RetentionPolicy, pointers: Vec<String>) -> BTreeSet<String> {
+    let mut split_pointers_by_name: BTreeMap<String, BTreeMap<u64, String>> = BTreeMap::new();
+
+    for pointer in pointers {
+        let (name, datetime) = if let Some(x) = split_pointer_name(&pointer) {
+            x
+        } else {
+            continue;
+        };
+
+        split_pointers_by_name
+            .entry(name)
+            .or_default()
+            .insert(datetime.timestamp().try_into().unwrap(), pointer);
+    }
+
+    let mut pointers_to_remove = BTreeSet::new();
+
+    for (_pointer_base_name, ts_to_pointer) in split_pointers_by_name {
+        let to_remove = policy.apply_returning_prunable(ts_to_pointer);
+
+        pointers_to_remove.extend(to_remove);
+    }
+
+    pointers_to_remove
+}
+
+#[cfg(test)]
+mod test {
+    use crate::commands::prune::{get_prunable_pointers, RetentionPolicy};
+    use crate::descriptor::RetentionPolicyConfig;
+
+    #[test]
+    fn test_prunable_pointers() {
+        let pointers = vec![
+            "alice+2022-09-28_05:00:00",
+            "alice+2022-09-28_02:00:00",
+            "alice+2022-09-21_05:00:00",
+            "alice+2022-09-14_05:00:00",
+            "alice+2022-09-08_05:00:00",
+            "alice+2022-09-07_05:00:00",
+            "alice+2022-09-01_05:00:00",
+            "bob+2022-09-28_06:00:00",
+            "bob+2022-09-28_03:00:00",
+            "bob+2022-09-21_06:00:00",
+            "bob+2022-09-14_06:00:00",
+            "bob+2022-09-08_06:00:00",
+            "bob+2022-09-07_06:00:00",
+            "bob+2022-09-01_06:00:00",
+        ]
+        .into_iter()
+        .map(|s| s.to_owned())
+        .collect();
+        let policy = RetentionPolicy::from_config(RetentionPolicyConfig {
+            daily: 0,
+            weekly: 3,
+            monthly: 0,
+            yearly: 0,
+        });
+
+        assert_eq!(
+            get_prunable_pointers(&policy, pointers)
+                .into_iter()
+                .collect::<Vec<_>>(),
+            vec![
+                "alice+2022-09-01_05:00:00",
+                "alice+2022-09-08_05:00:00",
+                "alice+2022-09-28_02:00:00",
+                "bob+2022-09-01_06:00:00",
+                "bob+2022-09-08_06:00:00",
+                "bob+2022-09-28_03:00:00",
+            ]
+        );
+    }
+}

datman/src/commands/pushpull.rs

@@ -1,9 +1,11 @@
 // Push and Pull support for Datman
 
 use anyhow::{bail, ensure, Context};
+use log::info;
 use std::collections::{BTreeMap, BTreeSet};
 use std::io::{Read, Write};
 use std::sync::Arc;
+use std::time::Instant;
 use yama::chunking::RecursiveUnchunker;
 use yama::commands::retrieve_tree_node;
 use yama::definitions::{ChunkId, PointerData, RecursiveChunkRef, TreeNode};
@@ -183,9 +185,25 @@ pub fn offering_side<R: Read, W: Write>(
     drop(chunks_to_offer);
     drop(chunks_to_skip);
 
-    progress.set_max_size(chunks_to_send.len() as u64);
+    let start_sort_by_hints = Instant::now();
+    let chunks_to_send_with_hints: BTreeSet<(u64, ChunkId)> = chunks_to_send
+        .into_iter()
+        .map(|chunk_id| {
+            pile.raw_pile
+                .chunk_id_transfer_ordering_hint(&chunk_id)
+                .map(|hint| (hint, chunk_id))
+        })
+        .collect::<anyhow::Result<_>>()?;
+    let time_to_sort_by_hints = Instant::now() - start_sort_by_hints;
+    info!(
+        "{} s to sort {} chunks by their hints",
+        time_to_sort_by_hints.as_secs_f32(),
+        chunks_to_send_with_hints.len()
+    );
+
+    progress.set_max_size(chunks_to_send_with_hints.len() as u64);
     progress.set_current(0);
-    for chunk_id in chunks_to_send {
+    for (_hint, chunk_id) in chunks_to_send_with_hints {
         let chunk_data = bypass_pile
             .read(Keyspace::Chunk, &chunk_id)?
             .context("Chunk vanished")?;

datman/src/descriptor.rs

@@ -38,6 +38,10 @@ pub struct Descriptor {
     pub piles: HashMap<String, DestPileDescriptor>,
 
     pub remote_hosts: HashMap<String, RemoteHostDescriptor>,
+
+    #[serde(default)]
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub retention: Option<RetentionPolicyConfig>,
 }
 
 #[derive(Clone, Serialize, Deserialize, Debug)]
@@ -46,6 +50,14 @@ pub struct RemoteHostDescriptor {
     pub path_to_datman: Option<String>,
 }
 
+#[derive(Clone, Serialize, Deserialize, Debug)]
+pub struct RetentionPolicyConfig {
+    pub daily: u32,
+    pub weekly: u32,
+    pub monthly: u32,
+    pub yearly: u32,
+}
+
 #[derive(Clone, Serialize, Deserialize, Debug)]
 #[serde(untagged)]
 pub enum SourceDescriptor {

datman/src/labelling.rs

@@ -15,10 +15,10 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
-use std::collections::HashMap;
+use std::collections::{BTreeSet, HashMap};
 use std::fs::File;
 use std::io::{BufRead, BufReader, Write};
-use std::path::Path;
+use std::path::{Path, PathBuf};
 
 use anyhow::anyhow;
 use anyhow::Context;
@@ -222,6 +222,23 @@ impl LabellingRules {
         }
         None
     }
+
+    pub fn get_exclusions_set(&self, base: &Path) -> BTreeSet<PathBuf> {
+        let mut exclusions = BTreeSet::new();
+
+        for (ext_path, state) in &self.position_based_rules {
+            assert!(ext_path.is_empty() || ext_path.starts_with('/'));
+            let full_path = PathBuf::from(format!(
+                "{}{ext_path}",
+                base.to_str().expect("base path must always be utf-8")
+            ));
+            if state == &Excluded {
+                exclusions.insert(full_path);
+            }
+        }
+
+        exclusions
+    }
 }
 
 /// Uninteractively label the nodes.

datman/src/remote/backup_source_requester.rs

@@ -1,11 +1,13 @@
 use crate::commands::backup::{get_pointer_name_at, label_filter_and_convert};
 use crate::descriptor::{Descriptor, DestPileDescriptor, SourceDescriptor};
+use crate::labelling::load_labelling_rules;
 use crate::tree::FileTree;
 use anyhow::{anyhow, bail};
 use chrono::Utc;
 use log::info;
+use std::collections::BTreeSet;
 use std::io::{Read, Write};
-use std::path::Path;
+use std::path::{Path, PathBuf};
 use std::process::{Child, Command, Stdio};
 use std::sync::Arc;
 use yama::commands::{load_pile_descriptor, open_pile};
@@ -48,11 +50,13 @@ pub fn scanning<R: Read, W: Write>(
     write: &mut W,
     path: &Path,
     one_filesystem: bool,
+    exclusions: &BTreeSet<PathBuf>,
 ) -> anyhow::Result<Option<FileTree<(), (), (), ()>>> {
     info!("Scanning.");
     write_message(write, &"scan")?;
     write_message(write, &path)?;
     write_message(write, &one_filesystem)?;
+    write_message(write, exclusions)?;
     write.flush()?;
     let scan_result: Option<FileTree<(), (), (), ()>> = read_message(read)?;
 
@@ -199,6 +203,9 @@ pub fn backup_remote_source_to_destination<PT: ProgressTracker + Send + 'static>
             info!("Connecting...");
             introduction(&mut read, &mut write)?;
 
+            let rules = load_labelling_rules(desc_path, source_name)?;
+            let exclusions = rules.get_exclusions_set(directory);
+
             // then request to scan
             info!("Requesting scan... (this may take some time)");
             let scan_result = scanning(
@@ -206,11 +213,12 @@ pub fn backup_remote_source_to_destination<PT: ProgressTracker + Send + 'static>
                 &mut write,
                 directory.as_ref(),
                 !*cross_filesystems,
+                &exclusions,
             )?
             .ok_or_else(|| anyhow!("Remote scan failed (does the directory exist?)"))?;
 
             let mut root =
-                label_filter_and_convert(scan_result, descriptor, desc_path, source_name, dest)?
+                label_filter_and_convert(scan_result, descriptor, source_name, &rules, dest)?
                     .ok_or_else(|| anyhow!("Empty filter..."))?;
 
             let absolute_dest_path = desc_path.join(&dest.path);

datman/src/remote/backup_source_responder.rs

@@ -1,6 +1,7 @@
 // This file implements the responder side of the backup source protocol -- the protocol used
 // to connect to remote backup sources.
 
+use std::collections::BTreeSet;
 use std::io::{stdin, stdout, Read, Write};
 use std::path::PathBuf;
 use std::sync::Arc;
@@ -43,7 +44,8 @@ pub fn introduction<R: Read, W: Write>(read: &mut R, write: &mut W) -> anyhow::R
 pub fn scanning<R: Read, W: Write>(read: &mut R, write: &mut W) -> anyhow::Result<()> {
     let path: PathBuf = read_message(read)?;
     let one_filesystem: bool = read_message(read)?;
-    let scan_result = scan(&path, one_filesystem)?;
+    let exclusions: BTreeSet<PathBuf> = read_message(read)?;
+    let scan_result = scan(&path, one_filesystem, &exclusions)?;
     write_message(write, &scan_result)?;
     write.flush()?;
     Ok(())

datman/src/tree.rs

@@ -15,16 +15,16 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
-use std::collections::BTreeMap;
+use std::collections::{BTreeMap, BTreeSet};
 use std::fmt::Debug;
 use std::fs::{read_link, symlink_metadata, DirEntry, Metadata};
 use std::io::ErrorKind;
 use std::os::unix::fs::MetadataExt;
-use std::path::Path;
+use std::path::{Path, PathBuf};
 
 use anyhow::anyhow;
 use indicatif::{ProgressBar, ProgressDrawTarget, ProgressStyle};
-use log::{info, warn};
+use log::{debug, info, warn};
 use serde::{Deserialize, Serialize};
 
 pub use yama::definitions::FilesystemOwnership;
@@ -216,14 +216,18 @@ pub fn mtime_msec(metadata: &Metadata) -> u64 {
 }
 
 /// Scan the filesystem to produce a Tree, using a default progress bar.
-pub fn scan(path: &Path, one_filesystem: bool) -> anyhow::Result<Option<FileTree<(), (), (), ()>>> {
+pub fn scan(
+    path: &Path,
+    one_filesystem: bool,
+    exclusions: &BTreeSet<PathBuf>,
+) -> anyhow::Result<Option<FileTree<(), (), (), ()>>> {
     let pbar = ProgressBar::with_draw_target(0, ProgressDrawTarget::stdout_with_hz(2));
     pbar.set_style(ProgressStyle::default_spinner().template("{spinner} {pos:7} {msg}"));
     pbar.set_message("dir scan");
 
     let one_filesystem = if one_filesystem { Some(None) } else { None };
 
-    let result = scan_with_progress_bar(path, &pbar, one_filesystem);
+    let result = scan_with_progress_bar(path, &pbar, one_filesystem, exclusions);
     pbar.finish_at_current_pos();
     result
 }
@@ -233,7 +237,14 @@ pub fn scan_with_progress_bar(
     path: &Path,
     progress_bar: &ProgressBar,
     mut one_filesystem: Option<Option<u64>>,
+    exclusions: &BTreeSet<PathBuf>,
 ) -> anyhow::Result<Option<FileTree<(), (), (), ()>>> {
+    if exclusions.contains(path) {
+        // Don't enter excluded paths.
+        debug!("Not descending into excluded path: {:?}", path);
+        return Ok(None);
+    }
+
     let metadata_res = symlink_metadata(path);
     progress_bar.inc(1);
     if let Err(e) = &metadata_res {
@@ -305,15 +316,23 @@ pub fn scan_with_progress_bar(
 
         for entry in dir_read? {
             let entry: DirEntry = entry?;
-            let scanned = scan_with_progress_bar(&entry.path(), progress_bar, one_filesystem)?;
+
+            if entry.file_name() == ".datmanskip" {
+                // Directories with .datmanskip in them are to be skipped entirely.
+                // TODO(perf): should this be checked upfront before some children may already
+                //     have been scanned?
+                debug!("Skipping {path:?} because it has a .datmanskip file.");
+                return Ok(None);
+            }
+
+            let scanned =
+                scan_with_progress_bar(&entry.path(), progress_bar, one_filesystem, exclusions)?;
             if let Some(scanned) = scanned {
-                children.insert(
-                    entry
-                        .file_name()
-                        .into_string()
-                        .expect("OsString not String"),
-                    scanned,
-                );
+                if let Ok(filename) = entry.file_name().into_string() {
+                    children.insert(filename, scanned);
+                } else {
+                    warn!("Non-UTF-8 filename; ignoring: {:?}", entry.file_name())
+                }
             }
         }
 

flake.lock

@@ -1,12 +1,15 @@
 {
   "nodes": {
     "flake-utils": {
+      "inputs": {
+        "systems": "systems"
+      },
       "locked": {
-        "lastModified": 1653893745,
-        "narHash": "sha256-0jntwV3Z8//YwuOjzhV2sgJJPt+HY6KhU7VZUL0fKZQ=",
+        "lastModified": 1710146030,
+        "narHash": "sha256-SZ5L6eA7HJ/nmkzGG7/ISclqe6oZdOZTNoesiInkXPQ=",
         "owner": "numtide",
         "repo": "flake-utils",
-        "rev": "1ed9fb1935d260de5fe1c2f7ee0ebaae17ed2fa1",
+        "rev": "b1d9ab70662946ef0850d488da1c9019f3a9752a",
         "type": "github"
       },
       "original": {
@@ -17,14 +20,16 @@
     },
     "naersk": {
       "inputs": {
-        "nixpkgs": "nixpkgs"
+        "nixpkgs": [
+          "nixpkgs"
+        ]
       },
       "locked": {
-        "lastModified": 1653413650,
-        "narHash": "sha256-wojDHjb+eU80MPH+3HQaK0liUy8EgR95rvmCl24i58Y=",
+        "lastModified": 1662220400,
+        "narHash": "sha256-9o2OGQqu4xyLZP9K6kNe1pTHnyPz0Wr3raGYnr9AIgY=",
         "owner": "nix-community",
         "repo": "naersk",
-        "rev": "69daaceebe12c070cd5ae69ba38f277bbf033695",
+        "rev": "6944160c19cb591eb85bbf9b2f2768a935623ed3",
         "type": "github"
       },
       "original": {
@@ -33,60 +38,58 @@
         "type": "github"
       }
     },
+    "nix-github-actions": {
+      "inputs": {
+        "nixpkgs": [
+          "poetry2nix",
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1703863825,
+        "narHash": "sha256-rXwqjtwiGKJheXB43ybM8NwWB8rO2dSRrEqes0S7F5Y=",
+        "owner": "nix-community",
+        "repo": "nix-github-actions",
+        "rev": "5163432afc817cf8bd1f031418d1869e4c9d5547",
+        "type": "github"
+      },
+      "original": {
+        "owner": "nix-community",
+        "repo": "nix-github-actions",
+        "type": "github"
+      }
+    },
     "nixpkgs": {
       "locked": {
-        "lastModified": 1653738054,
-        "narHash": "sha256-IaR8iLN4Ms3f5EjU1CJkXSc49ZzyS5qv03DtVAti6/s=",
+        "lastModified": 1714971268,
+        "narHash": "sha256-IKwMSwHj9+ec660l+I4tki/1NRoeGpyA2GdtdYpAgEw=",
         "owner": "NixOS",
         "repo": "nixpkgs",
-        "rev": "17b62c338f2a0862a58bb6951556beecd98ccda9",
+        "rev": "27c13997bf450a01219899f5a83bd6ffbfc70d3c",
         "type": "github"
       },
       "original": {
         "id": "nixpkgs",
+        "ref": "nixos-23.11",
         "type": "indirect"
       }
     },
-    "nixpkgs_2": {
-      "locked": {
-        "lastModified": 1653738054,
-        "narHash": "sha256-IaR8iLN4Ms3f5EjU1CJkXSc49ZzyS5qv03DtVAti6/s=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "17b62c338f2a0862a58bb6951556beecd98ccda9",
-        "type": "github"
-      },
-      "original": {
-        "id": "nixpkgs",
-        "type": "indirect"
-      }
-    },
-    "nixpkgs_3": {
-      "locked": {
-        "lastModified": 1654176133,
-        "narHash": "sha256-XhjUlU+q9LPFM8Z7X3h504vS2FUToCyKhaf5hVF6nsw=",
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "rev": "63f15db5291aec276924d907d3e083e74d68e8b9",
-        "type": "github"
-      },
-      "original": {
-        "owner": "NixOS",
-        "repo": "nixpkgs",
-        "type": "github"
-      }
-    },
     "poetry2nix": {
       "inputs": {
         "flake-utils": "flake-utils",
-        "nixpkgs": "nixpkgs_3"
+        "nix-github-actions": "nix-github-actions",
+        "nixpkgs": [
+          "nixpkgs"
+        ],
+        "systems": "systems_2",
+        "treefmt-nix": "treefmt-nix"
       },
       "locked": {
-        "lastModified": 1653561148,
-        "narHash": "sha256-JzAttqACdvMOTwkzkJ0jFF8MWIo8Uau4w/XUMyqpnd8=",
+        "lastModified": 1715017507,
+        "narHash": "sha256-RN2Vsba56PfX02DunWcZYkMLsipp928h+LVAWMYmbZg=",
         "owner": "nix-community",
         "repo": "poetry2nix",
-        "rev": "3b01c3e3dc57d511848d8433153ab67db79640e1",
+        "rev": "e6b36523407ae6a7a4dfe29770c30b3a3563b43a",
         "type": "github"
       },
       "original": {
@@ -98,18 +101,68 @@
     "root": {
       "inputs": {
         "naersk": "naersk",
-        "nixpkgs": "nixpkgs_2",
+        "nixpkgs": "nixpkgs",
         "poetry2nix": "poetry2nix",
         "utils": "utils"
       }
     },
+    "systems": {
+      "locked": {
+        "lastModified": 1681028828,
+        "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=",
+        "owner": "nix-systems",
+        "repo": "default",
+        "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e",
+        "type": "github"
+      },
+      "original": {
+        "owner": "nix-systems",
+        "repo": "default",
+        "type": "github"
+      }
+    },
+    "systems_2": {
+      "locked": {
+        "lastModified": 1681028828,
+        "narHash": "sha256-Vy1rq5AaRuLzOxct8nz4T6wlgyUR7zLU309k9mBC768=",
+        "owner": "nix-systems",
+        "repo": "default",
+        "rev": "da67096a3b9bf56a91d16901293e51ba5b49a27e",
+        "type": "github"
+      },
+      "original": {
+        "id": "systems",
+        "type": "indirect"
+      }
+    },
+    "treefmt-nix": {
+      "inputs": {
+        "nixpkgs": [
+          "poetry2nix",
+          "nixpkgs"
+        ]
+      },
+      "locked": {
+        "lastModified": 1714058656,
+        "narHash": "sha256-Qv4RBm4LKuO4fNOfx9wl40W2rBbv5u5m+whxRYUMiaA=",
+        "owner": "numtide",
+        "repo": "treefmt-nix",
+        "rev": "c6aaf729f34a36c445618580a9f95a48f5e4e03f",
+        "type": "github"
+      },
+      "original": {
+        "owner": "numtide",
+        "repo": "treefmt-nix",
+        "type": "github"
+      }
+    },
     "utils": {
       "locked": {
-        "lastModified": 1652776076,
-        "narHash": "sha256-gzTw/v1vj4dOVbpBSJX4J0DwUR6LIyXo7/SuuTJp1kM=",
+        "lastModified": 1659877975,
+        "narHash": "sha256-zllb8aq3YO3h8B/U0/J1WBgAL8EX5yWf5pMj3G0NAmc=",
         "owner": "numtide",
         "repo": "flake-utils",
-        "rev": "04c1b180862888302ddfb2e3ad9eaa63afc60cf8",
+        "rev": "c0e246b9b83f637f4681389ecabcb2681b4f3af0",
         "type": "github"
       },
       "original": {

flake.nix

@@ -3,13 +3,21 @@
 
   inputs = {
     utils.url = "github:numtide/flake-utils";
-    naersk.url = "github:nix-community/naersk";
-    poetry2nix.url = "github:nix-community/poetry2nix";
+    naersk = {
+      url = "github:nix-community/naersk";
+      inputs.nixpkgs.follows = "nixpkgs";
+    };
+    nixpkgs.url = "nixpkgs/nixos-23.11";
+    poetry2nix = {
+      url = "github:nix-community/poetry2nix";
+      inputs.nixpkgs.follows = "nixpkgs";
+    };
   };
 
   outputs = { self, nixpkgs, utils, naersk, poetry2nix }:
     utils.lib.eachDefaultSystem (system: let
       pkgs = nixpkgs.legacyPackages."${system}";
+      inherit (poetry2nix.lib.mkPoetry2Nix { inherit pkgs; }) mkPoetryApplication;
       naersk-lib = naersk.lib."${system}";
 
       rustComponents = naersk-lib.buildPackage {
@@ -18,16 +26,16 @@
 
         buildInputs = with pkgs; [
           openssl
-          pkgconfig
+          pkg-config
           sqlite
         ];
       };
 
-      mysqlHelper = pkgs.poetry2nix.mkPoetryApplication {
+      mysqlHelper = mkPoetryApplication {
         projectDir = ./datman-helper-mysql;
       };
 
-      postgresHelper = pkgs.poetry2nix.mkPoetryApplication {
+      postgresHelper = mkPoetryApplication {
         projectDir = ./datman-helper-postgres;
       };
 

scripts-dev/lint.sh

@@ -4,7 +4,7 @@ if [ $# -ge 1 ]
 then
   files=$*
 else
-  files="testsuite/setup.py testsuite/datmantests testsuite/helpers testsuite/yamatests datman-helper-postgres/datman_helper_postgres datman-helper-postgres/setup.py datman-helper-mysql/datman_helper_mysql datman-helper-mysql/setup.py"
+  files="testsuite/setup.py testsuite/datmantests testsuite/helpers testsuite/yamatests datman-helper-postgres/datman_helper_postgres datman-helper-mysql/datman_helper_mysql"
 fi
 
 echo "Linting these locations: $files"

shell.nix

@@ -0,0 +1,50 @@
+{ pkgs ? import <nixpkgs> {} }:
+
+let
+  # We may need some packages from nixpkgs-unstable
+  #unstable = import <nixpkgs-unstable> {};
+
+  rust-toolchain = pkgs.symlinkJoin {
+    name = "rust-toolchain";
+    paths = [pkgs.rustc pkgs.cargo pkgs.rustfmt pkgs.rustPlatform.rustcSrc];
+  };
+in
+
+pkgs.mkShell {
+
+  buildInputs = [
+    rust-toolchain
+
+    pkgs.pkg-config
+
+    pkgs.alsa-lib
+    pkgs.sqlite
+    #pkgs.libclang # ??
+  ];
+
+  nativeBuildInputs = [
+    pkgs.openssl
+    pkgs.python3
+  ];
+
+  # Needed for bindgen when binding to avahi
+  LIBCLANG_PATH="${pkgs.llvmPackages_latest.libclang.lib}/lib";
+
+  # Cargo culted:
+  # Add to rustc search path
+  RUSTFLAGS = (builtins.map (a: ''-L ${a}/lib'') [
+  ]);
+  # Add to bindgen search path
+  BINDGEN_EXTRA_CLANG_ARGS =
+    # Includes with normal include path
+    (builtins.map (a: ''-I"${a}/include"'') [
+    ])
+    # Includes with special directory paths
+    ++ [
+      ''-I"${pkgs.llvmPackages_latest.libclang.lib}/lib/clang/${pkgs.llvmPackages_latest.libclang.version}/include"''
+      #''-I"${pkgs.glib.dev}/include/glib-2.0"''
+      #''-I${pkgs.glib.out}/lib/glib-2.0/include/''
+    ];
+
+
+}

testsuite/datmantests/test_backup_and_extract.py

@@ -251,7 +251,8 @@ kind = {{ stdout = "blahblah.txt" }}
         seed = 7555
         print(f"seed: {seed}")
         rng.seed(seed)
-        # min_files is 8 because we need enough files to use each label for this test to succeed.
+        # min_files is 8 because we need enough files to use each label for this
+        # test to succeed.
         initial_descriptor, _ = generate_random_dir(rng, src_path, 32, min_files=8)
         labellings = generate_labels(initial_descriptor, rng)
         save_labelling_rules(labelling_path.joinpath("srca.zst"), labellings)
@@ -298,3 +299,81 @@ kind = {{ stdout = "blahblah.txt" }}
             )
 
         td.cleanup()
+
+    def test_backup_incremental_with_mid_delete(self):
+        td = TemporaryDirectory("test_backup_incremental_with_mid_delete")
+        tdpath = Path(td.name)
+
+        datman_path = tdpath.joinpath("datman")
+        src_path = datman_path.joinpath("srca")
+        yama_path = datman_path.joinpath("main")
+
+        set_up_simple_datman(datman_path)
+        set_up_simple_yama(yama_path)
+
+        rng = Random()
+        seed = rng.randint(0, 9001)
+        print(f"seed: {seed}")
+        rng.seed(seed)
+        initial_descriptor, _ = generate_random_dir(rng, src_path, 32)
+
+        print("storing")
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+
+        # now mutate and store incremental
+        randomly_mutate_directory_in_descriptor(initial_descriptor, src_path, rng)
+        time.sleep(2)
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+
+        # now mutate and store incremental again!
+        randomly_mutate_directory_in_descriptor(initial_descriptor, src_path, rng)
+        mutated_descriptor = scan_dir(src_path)
+        time.sleep(2)
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+
+        pointer_names = [
+            line
+            for line in subprocess.check_output(("yama", "debug", "lsp"), cwd=yama_path)
+            .decode()
+            .split("\n")
+            if line
+        ]
+        self.assertEqual(len(pointer_names), 3)
+        self.assertLess(pointer_names[0], pointer_names[1])
+        self.assertLess(pointer_names[1], pointer_names[2])
+
+        print(f"removing mid pointer {pointer_names[1]}")
+        subprocess.check_call(
+            ("yama", "debug", "rmp", pointer_names[1]),
+            cwd=yama_path,
+        )
+
+        print("extracting last pointer to check still valid")
+        dest_path = tdpath.joinpath("desta")
+        subprocess.check_call(
+            (
+                "datman",
+                "extract",
+                "--skip-metadata",
+                "--accept-partial",
+                "main",
+                "../desta",
+            ),
+            cwd=datman_path,
+        )
+
+        # this will be wrapped in a directory that starts with the name srca+
+        extracted_dir_descriptor_wrapper = scan_dir(dest_path)
+
+        contents = extracted_dir_descriptor_wrapper.contents
+        self.assertEqual(len(contents), 1)
+        key, value = next(iter(contents.items()))
+        self.assertTrue(key.startswith("srca+"))
+
+        self.assertIsInstance(value, DirectoryDescriptor)
+        key, value = next(iter(value.contents.items()))
+        self.assertEqual(key, "srca")
+
+        self.assertEqual(value.ignore_metadata(), mutated_descriptor.ignore_metadata())
+
+        td.cleanup()

testsuite/helpers/yama_helpers.py

@@ -1,6 +1,7 @@
 import shutil
 import subprocess
 from pathlib import Path
+from typing import Set
 
 
 def set_up_simple_yama(path: Path):
@@ -10,3 +11,13 @@ def set_up_simple_yama(path: Path):
         "example_zstd.dict"
     )
     shutil.copyfile(example_zstd_path, path.joinpath("important_zstd.dict"))
+
+
+def list_bloblog_ids(pile: Path) -> Set[int]:
+    result = set()
+    for p in pile.joinpath("bloblog").iterdir():
+        try:
+            result.add(int(p.name))
+        except ValueError:
+            pass
+    return result

testsuite/setup.py

@@ -22,7 +22,7 @@ REQUIRED = ["green", "attrs", "immutabledict"]
 
 
 # What packages are optional?
-EXTRAS = {"dev": ["black==21.7b0", "flake8==3.9.2", "isort==5.9.2"]}
+EXTRAS = {"dev": ["black==22.10.0", "flake8==3.9.2", "isort==5.9.2"]}
 
 # The rest you shouldn't have to touch too much :)
 # ------------------------------------------------

testsuite/yamatests/test_compact.py

@@ -0,0 +1,175 @@
+import subprocess
+from pathlib import Path
+from random import Random
+from tempfile import TemporaryDirectory
+from unittest import TestCase
+
+from helpers import (
+    DirectoryDescriptor,
+    generate_random_dir,
+    randomly_mutate_directory_in_descriptor,
+    scan_dir,
+)
+from helpers.datman_helpers import set_up_simple_datman
+from helpers.yama_helpers import list_bloblog_ids, set_up_simple_yama
+
+
+class TestYamaCompact(TestCase):
+    def test_compaction_merge_two_small_bloblogs(self):
+        td = TemporaryDirectory("test_check_fails_after_random_corruption")
+        tdpath = Path(td.name)
+
+        datman_path = tdpath.joinpath("datman")
+        src_path = datman_path.joinpath("srca")
+        yama_path = datman_path.joinpath("main")
+
+        set_up_simple_datman(datman_path)
+        set_up_simple_yama(yama_path)
+
+        rng = Random()
+        seed = rng.randint(0, 9001)
+        print(f"seed: {seed}")
+        rng.seed(seed)
+        later_expected_descriptor, _ = generate_random_dir(rng, src_path, 32)
+
+        # Back up twice: that way we should get at least two bloblogs!
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+        old_bloblog_ids = list_bloblog_ids(yama_path)
+        self.assertGreater(
+            len(old_bloblog_ids), 1, "Should be many bloblogs at this point"
+        )
+
+        subprocess.check_call(
+            (
+                "yama",
+                "compact",
+                "--mergeable",
+                "2",
+                "--small",
+                str(2 * 1024 * 1024 * 1024),
+            ),
+            cwd=yama_path,
+        )
+
+        new_bloblog_ids = list_bloblog_ids(yama_path)
+        self.assertEqual(
+            len(new_bloblog_ids), 1, "Should only be 1 bloblog at this point."
+        )
+        self.assertEqual(
+            list(new_bloblog_ids)[0],
+            max(old_bloblog_ids) + 1,
+            "New bloblog ID should be 1 greater than the max old one.",
+        )
+
+    def test_gc_then_compact(self):
+        td = TemporaryDirectory("test_gc_then_compact")
+        tdpath = Path(td.name)
+
+        datman_path = tdpath.joinpath("datman")
+        src_path = datman_path.joinpath("srca")
+        yama_path = datman_path.joinpath("main")
+
+        set_up_simple_datman(datman_path)
+        set_up_simple_yama(yama_path)
+
+        rng = Random()
+        seed = rng.randint(0, 9001)
+        print(f"seed: {seed}")
+        rng.seed(seed)
+        initial_descriptor, _ = generate_random_dir(rng, src_path, 32)
+
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+        orig_pointer_name = (
+            subprocess.check_output(("yama", "debug", "lsp"), cwd=yama_path)
+            .decode()
+            .split("\n")[0]
+        )
+
+        randomly_mutate_directory_in_descriptor(initial_descriptor, src_path, rng)
+        mutated_descriptor = scan_dir(src_path)
+
+        subprocess.check_call(("datman", "backup-one", "srca", "main"), cwd=datman_path)
+
+        old_bloblog_ids = list_bloblog_ids(yama_path)
+
+        # Try a GC and check that it's a no-op
+        subprocess.check_call(
+            ("yama", "check", "--shallow", "--apply-gc"), cwd=yama_path
+        )
+        subprocess.check_call(
+            (
+                "yama",
+                "compact",
+                "--mergeable",
+                "2000",
+                "--reclaim",
+                "1",
+                "--max-dealloc",
+                "1",
+            ),
+            cwd=yama_path,
+        )
+
+        unchanged_bloblog_ids = list_bloblog_ids(yama_path)
+        self.assertEqual(
+            old_bloblog_ids,
+            unchanged_bloblog_ids,
+            "No GC: no compaction should have happened.",
+        )
+
+        subprocess.check_call(
+            ("yama", "debug", "rmp", orig_pointer_name), cwd=yama_path
+        )
+
+        # Try a GC and check that it did something
+        subprocess.check_call(
+            ("yama", "check", "--shallow", "--apply-gc"), cwd=yama_path
+        )
+        subprocess.check_call(
+            (
+                "yama",
+                "compact",
+                "--mergeable",
+                "2000",
+                "--reclaim",
+                "1",
+                "--max-dealloc",
+                "1",
+            ),
+            cwd=yama_path,
+        )
+
+        new_bloblog_ids = list_bloblog_ids(yama_path)
+        self.assertNotEqual(
+            old_bloblog_ids, new_bloblog_ids, "GC: compaction should have happened."
+        )
+
+        # Check that we can still extract the files!
+        dest_path = tdpath.joinpath("desta")
+        subprocess.check_call(
+            (
+                "datman",
+                "extract",
+                "--skip-metadata",
+                "--accept-partial",
+                "main",
+                "../desta",
+            ),
+            cwd=datman_path,
+        )
+
+        extracted_dir_descriptor_wrapper = scan_dir(dest_path)
+
+        contents = extracted_dir_descriptor_wrapper.contents
+        self.assertEqual(len(contents), 1)
+        key, value = next(iter(contents.items()))
+        self.assertTrue(key.startswith("srca+"))
+
+        self.assertIsInstance(value, DirectoryDescriptor)
+        key, value = next(iter(value.contents.items()))
+        self.assertEqual(key, "srca")
+
+        self.assertEqual(value.ignore_metadata(), mutated_descriptor.ignore_metadata())
+
+        td.cleanup()

yama/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "yama"
-version = "0.6.0-alpha.4"
+version = "0.6.0-alpha.5"
 authors = ["Olivier 'reivilibre' <olivier@librepush.net>"]
 edition = "2018"
 description = "Deduplicated, compressed and encrypted content pile manager"

yama/src/bin/yama.rs

@@ -29,7 +29,8 @@ use yama::operations::checking::VacuumMode;
 use yama::operations::legacy_pushpull::{
     determine_bypass_level, open_pile_with_work_bypass, push_to,
 };
-use yama::operations::{checking, extracting};
+use yama::operations::{checking, cleanup, extracting};
+use yama::pile::local_sqlitebloblogs::CompactionThresholds;
 use yama::pile::{Pile, PileDescriptor, RawPile};
 use yama::{commands, debug};
 
@@ -56,8 +57,9 @@ enum PileCommand {
         pointer_name: String,
 
         /// Limited expression(s) of files to retrieve.
+        /// LIMITATION OF CURRENT VERSION: ONLY ONE EXACT PATH ALLOWED, PLEASE.
         #[clap(short, long)]
-        subset: Vec<PathBuf>,
+        subset: Option<String>,
 
         destination: PathBuf,
 
@@ -82,6 +84,29 @@ enum PileCommand {
         shallow: bool,
     },
 
+    Compact {
+        /// Don't actually perform any compaction; just plan it out.
+        #[clap(long)]
+        dry_run: bool,
+
+        /// Allocated size under which a bloblog is considered small.
+        #[clap(long = "small")]
+        small_thresh: Option<u64>,
+
+        /// Minimum amount of space to reclaim in order to run compaction for reclaim.
+        #[clap(long = "reclaim")]
+        min_reclaim: Option<u64>,
+
+        /// Maximum amount of space that can be deallocated in a bloblog before we consider it
+        /// worthwhile to replace.
+        #[clap(long = "max-dealloc")]
+        max_deallocated: Option<u64>,
+
+        /// Minimum number of mergeable small bloblogs in order to run compaction for merge.
+        #[clap(long)]
+        mergeable: Option<u32>,
+    },
+
     /// Enter a debug prompt for manually operating on the yama pile.
     Debug { supplied_command: Vec<String> },
 
@@ -118,7 +143,7 @@ fn wrapped_main() -> anyhow::Result<i32> {
     match &opts.command {
         PileCommand::Retrieve {
             pointer_name,
-            subset: _,
+            subset,
             destination,
             num_workers: workers,
         } => {
@@ -136,10 +161,25 @@ fn wrapped_main() -> anyhow::Result<i32> {
 
             fully_integrate_pointer_node(&pile, &mut root_tree_node.node, &mut pointer)?;
 
+            let mut node_to_extract = &mut root_tree_node.node;
+
+            if let Some(subset) = subset {
+                for path_to_descend in subset.split('/').filter(|s| !s.is_empty()) {
+                    match node_to_extract.child(path_to_descend) {
+                        Ok(new_node) => {
+                            node_to_extract = new_node;
+                        }
+                        Err(msg) => {
+                            bail!("Can't descend into {path_to_descend:?}: {msg}");
+                        }
+                    }
+                }
+            }
+
             // todo allow disabling apply metadata
             extracting::extract(
                 destination,
-                &mut root_tree_node.node,
+                node_to_extract,
                 &pile,
                 true,
                 workers.unwrap_or(2),
@@ -175,6 +215,29 @@ fn wrapped_main() -> anyhow::Result<i32> {
                 return Ok(1);
             }
         }
+        PileCommand::Compact {
+            dry_run,
+            small_thresh,
+            min_reclaim,
+            max_deallocated,
+            mergeable,
+        } => {
+            let this_dir = Path::new(".");
+            let descriptor =
+                load_pile_descriptor(this_dir).context("Failed to load pile descriptor")?;
+            cleanup::compact(
+                this_dir,
+                &descriptor,
+                !*dry_run,
+                true,
+                CompactionThresholds {
+                    minimum_to_reclaim: min_reclaim.unwrap_or(2 * 1024 * 1024 * 1024),
+                    minimum_small_bloblogs_to_merge: mergeable.unwrap_or(64),
+                    cond_if_more_deallocated_than: max_deallocated.unwrap_or(256 * 1024 * 1024),
+                    cond_if_less_allocated_than: small_thresh.unwrap_or(64 * 1024 * 1024),
+                },
+            )?;
+        }
         PileCommand::Init {} => {
             commands::init(".".as_ref())?;
         }

yama/src/debug.rs

@@ -15,13 +15,12 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
-use crate::commands::{fully_integrate_pointer_node, retrieve_tree_node, store_tree_node};
+use crate::commands::retrieve_tree_node;
 use crate::definitions::{FilesystemOwnership, FilesystemPermissions, TreeNode};
+use crate::operations::remove_pointer_safely;
 use crate::pile::{Pile, PileDescriptor, RawPile};
-use crate::tree::integrate_node_in_place;
 use anyhow::anyhow;
 use clap::Parser;
-use log::info;
 use rustyline::error::ReadlineError;
 use rustyline::Editor;
 
@@ -123,50 +122,7 @@ pub fn debug_command<RP: RawPile>(
             }
         }
         DebugCommand::DeletePointer { name } => {
-            // retrieve this pointer
-            let mut this_pointer = pile.read_pointer(name.as_str())?.ok_or_else(|| {
-                anyhow!("Pointer {:?} does not exist so can not be deleted.", name)
-            })?;
-            let mut this_node = retrieve_tree_node(&pile, this_pointer.chunk_ref.clone())?;
-
-            // fully integrate the pointer
-            fully_integrate_pointer_node(&pile, &mut this_node.node, &mut this_pointer)?;
-            assert!(this_pointer.parent_pointer.is_none());
-
-            // now integrate any pointers that rely on this one
-            // so that they no longer rely on this one.
-            for pointer in pile.list_pointers()?.iter() {
-                if pointer == name {
-                    continue;
-                }
-                if let Some(mut pointer_data) = pile.read_pointer(pointer.as_str())? {
-                    if let Some(parent_pointer) = pointer_data.parent_pointer.as_ref() {
-                        if parent_pointer == name {
-                            info!("Pointer is now an orphan: {:?}", pointer);
-
-                            // need to integrate this node, so retrieve it
-                            let mut node = retrieve_tree_node(&pile, pointer_data.chunk_ref)?;
-
-                            // integrate it in-place
-                            integrate_node_in_place(&mut node.node, &this_node.node)?;
-
-                            // mark it as orphaned (no parent)
-                            pointer_data.parent_pointer = None;
-
-                            // store the orphaned node
-                            let new_chunk_ref = store_tree_node(&pile, &node)?;
-                            // associate the orphaned node with the orphaned pointer
-                            pointer_data.chunk_ref = new_chunk_ref;
-                            // write the pointer back.
-                            pile.write_pointer(pointer.as_str(), &pointer_data)?;
-                        }
-                    }
-                }
-            }
-
-            // then delete the pointer
-            pile.delete_pointer(name)?;
-            info!("Deleted pointer: {:?}", name);
+            remove_pointer_safely(pile, name)?;
         }
         DebugCommand::PointerInfo { name } => {
             let this_pointer = pile

yama/src/definitions.rs

@@ -270,6 +270,19 @@ impl TreeNode {
             }
         }
     }
+
+    /// Recurses into a child by name, or returns Err with a reason.
+    pub fn child(&mut self, name: &str) -> Result<&mut TreeNode, &'static str> {
+        match self {
+            TreeNode::NormalFile { .. } => Err("not a directory: normal file"),
+            TreeNode::Directory { children, .. } => match children.get_mut(name) {
+                None => Err("child not in directory"),
+                Some(node) => Ok(node),
+            },
+            TreeNode::SymbolicLink { .. } => Err("not a directory: symlink"),
+            TreeNode::Deleted => Err("not a directory: deleted"),
+        }
+    }
 }
 
 #[derive(Debug, Copy, Clone, Serialize, Deserialize, PartialEq, Eq)]

yama/src/operations.rs

@@ -1,4 +1,80 @@
+use crate::commands::{fully_integrate_pointer_node, retrieve_tree_node, store_tree_node};
+use crate::pile::{Pile, RawPile};
+use crate::tree::{differentiate_node_in_place, integrate_node_in_place};
+use anyhow::{anyhow, Context};
+use log::info;
+
 pub mod checking;
+pub mod cleanup;
 pub mod extracting;
 pub mod legacy_pushpull;
 pub mod storing;
+
+pub fn remove_pointer_safely<P: RawPile>(pile: &Pile<P>, name: &str) -> anyhow::Result<()> {
+    // retrieve this pointer
+    let mut this_pointer = pile
+        .read_pointer(name)?
+        .ok_or_else(|| anyhow!("Pointer {:?} does not exist so can not be deleted.", name))?;
+    let mut this_node = retrieve_tree_node(&pile, this_pointer.chunk_ref.clone())
+        .context("retrieving 'this' node")?;
+
+    let new_parent_name = this_pointer.parent_pointer.clone();
+    fully_integrate_pointer_node(pile, &mut this_node.node, &mut this_pointer)
+        .context("integrating new parent")?;
+
+    let new_parent = if let Some(ref new_parent_name) = new_parent_name {
+        let mut new_parent_pointer = pile
+            .read_pointer(new_parent_name.as_str())?
+            .ok_or_else(|| anyhow!("Parent pointer {:?} does not exist.", name))?;
+        let mut new_parent_node = retrieve_tree_node(&pile, new_parent_pointer.chunk_ref.clone())?;
+        fully_integrate_pointer_node(pile, &mut new_parent_node.node, &mut new_parent_pointer)?;
+        Some((new_parent_pointer, new_parent_node))
+    } else {
+        None
+    };
+
+    // now integrate any pointers that rely on this one
+    // so that they no longer rely on this one.
+    for pointer in pile.list_pointers()?.iter() {
+        if pointer == name {
+            continue;
+        }
+        if let Some(mut pointer_data) = pile.read_pointer(pointer.as_str())? {
+            if let Some(parent_pointer) = pointer_data.parent_pointer.as_ref() {
+                if parent_pointer == name {
+                    info!("Pointer would be orphaned: {:?}; integrating", pointer);
+
+                    // need to integrate this node, so retrieve it
+                    let mut node = retrieve_tree_node(&pile, pointer_data.chunk_ref)?;
+
+                    // integrate it in-place
+                    integrate_node_in_place(&mut node.node, &this_node.node)?;
+
+                    if let Some((_, ref new_parent_node)) = new_parent {
+                        // then differentiate with respect to the NEW parent
+                        differentiate_node_in_place(&mut node.node, &new_parent_node.node)?;
+                    }
+
+                    // pass through the parent
+                    pointer_data.parent_pointer = new_parent_name.clone();
+
+                    // store the updated version of the pointer
+                    let new_chunk_ref = store_tree_node(&pile, &node)?;
+                    // associate the new node with the new version of the pointer
+                    pointer_data.chunk_ref = new_chunk_ref;
+                    // write the pointer back.
+                    pile.write_pointer(pointer.as_str(), &pointer_data)?;
+
+                    // we must flush chunks before deleting the pointer
+                    pile.flush()
+                        .context("flushing after writing pointer back")?;
+                }
+            }
+        }
+    }
+
+    // then delete the pointer
+    pile.delete_pointer(name)?;
+    info!("Deleted pointer: {:?}", name);
+    Ok(())
+}

yama/src/operations/checking.rs

@@ -24,8 +24,10 @@ use crate::pile::{
 use anyhow::bail;
 use crossbeam_channel::Sender;
 use indicatif::{ProgressBar, ProgressDrawTarget, ProgressStyle};
+use itertools::Itertools;
 use log::{error, info, warn};
 use std::collections::HashSet;
+use std::convert::TryInto;
 use std::io::{Read, Write};
 use std::sync::Mutex;
 
@@ -110,6 +112,10 @@ impl<RP: RawPile> RawPile for VacuumRawPile<RP> {
         self.underlying.delete(kind, key)
     }
 
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        self.underlying.delete_many(kind, keys)
+    }
+
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -137,6 +143,10 @@ impl<RP: RawPile> RawPile for VacuumRawPile<RP> {
     fn describe_pipeline(&self) -> anyhow::Result<Vec<PipelineDescription>> {
         self.underlying.describe_pipeline()
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.underlying.chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }
 
 /// Runs a full check of a Yama pile. This reads ALL the chunks, which can take a long time.
@@ -403,9 +413,21 @@ pub fn check_shallow<RP: RawPile>(
 
             // actually do the vacuum!
             info!("Going to vacuum them up.");
-            for vacuum_id in to_vacuum {
-                pile.raw_pile.delete(Keyspace::Chunk, &vacuum_id)?;
-                pbar.inc(1);
+            for vacuum_ids_chunk in to_vacuum
+                .into_iter()
+                .chunks(512)
+                .into_iter()
+                .map(|c| c.collect::<Vec<ChunkId>>())
+            {
+                pile.raw_pile.delete_many(
+                    Keyspace::Chunk,
+                    vacuum_ids_chunk
+                        .iter()
+                        .map(|ci| ci.as_slice())
+                        .collect::<Vec<&[u8]>>()
+                        .as_slice(),
+                )?;
+                pbar.inc(vacuum_ids_chunk.len().try_into().unwrap());
             }
             pile.flush()?;
             pbar.finish_and_clear();

yama/src/operations/cleanup.rs

@@ -0,0 +1,64 @@
+use crate::pile::local_sqlitebloblogs::{CompactionThresholds, SqliteBloblogPile};
+use crate::pile::{PileDescriptor, PileStorage};
+use anyhow::{bail, Context};
+use indicatif::{ProgressBar, ProgressDrawTarget, ProgressStyle};
+use log::info;
+use std::path::Path;
+
+pub fn compact(
+    pile_path: &Path,
+    pile_desc: &PileDescriptor,
+    actually_run: bool,
+    make_progress_bar: bool,
+    thresholds: CompactionThresholds,
+) -> anyhow::Result<()> {
+    let pbar = if make_progress_bar {
+        ProgressBar::with_draw_target(1000 as u64, ProgressDrawTarget::stdout_with_hz(10))
+    } else {
+        ProgressBar::hidden()
+    };
+    pbar.set_style(
+        ProgressStyle::default_bar()
+            .template("[{elapsed_precise}]/[{eta}] {bar:40.cyan/blue} {pos:>7}/{len:7} {msg}"),
+    );
+    pbar.set_message("compacting");
+
+    match pile_desc.storage {
+        PileStorage::SqliteIndexedBloblog => {
+            let bloblog_pile = SqliteBloblogPile::open(&pile_path)
+                .context("Failed to open SQLite-indexed Bloblog Pile")?;
+            compact_bloblogs(bloblog_pile, pbar, actually_run, thresholds)?;
+            Ok(())
+        }
+        other @ PileStorage::RemoteOnly => {
+            bail!("Cannot use compaction on this kind of pile: {other:?}!");
+        }
+    }
+}
+
+fn compact_bloblogs(
+    bloblog_pile: SqliteBloblogPile,
+    pbar: ProgressBar,
+    actually_run: bool,
+    thresholds: CompactionThresholds,
+) -> anyhow::Result<()> {
+    info!("=== Analysing for compaction ===");
+    let analysis = bloblog_pile.analyse_for_compaction()?;
+    let chunks_total: u64 = analysis.values().map(|bs| bs.chunks_total).sum();
+    let chunks_deleted: u64 = analysis.values().map(|bs| bs.chunks_deleted).sum();
+    let bytes_total: u64 = analysis.values().map(|bs| bs.bytes_total).sum();
+    let bytes_deleted: u64 = analysis.values().map(|bs| bs.bytes_deleted).sum();
+
+    info!("{} bloblogs in this pile, with {chunks_total} chunks ({bytes_total} B) of which {chunks_deleted} ({bytes_deleted} B) are deleted.", analysis.len());
+
+    info!("=== Planning compaction ===");
+    let plan = bloblog_pile.plan_compaction(&thresholds, analysis)?;
+    info!("Planned compaction: replace {} bloblogs (of which {} are small), freeing up {} B and rewriting {} B", plan.bloblogs_to_replace.len(), plan.small_bloblogs, plan.reclaimable_space, plan.bytes_to_write);
+
+    if actually_run {
+        info!("=== Compacting ===");
+        bloblog_pile.perform_compaction(Box::new(pbar), plan)?;
+    }
+
+    Ok(())
+}

yama/src/pile.rs

@@ -141,6 +141,7 @@ pub trait RawPile: Send + Sync + Debug + 'static {
     fn read(&self, kind: Keyspace, key: &[u8]) -> anyhow::Result<Option<Vec<u8>>>;
     fn write(&self, kind: Keyspace, key: &[u8], value: &[u8]) -> anyhow::Result<()>;
     fn delete(&self, kind: Keyspace, key: &[u8]) -> anyhow::Result<()>;
+    fn delete_many(&self, kind: Keyspace, key: &[&[u8]]) -> anyhow::Result<()>;
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -167,6 +168,10 @@ pub trait RawPile: Send + Sync + Debug + 'static {
     ) -> anyhow::Result<Sender<(ChunkId, Vec<u8>)>>;
 
     fn describe_pipeline(&self) -> anyhow::Result<Vec<PipelineDescription>>;
+
+    /// Return a u64 order token that indicates the optimum order to read this chunk in
+    /// compared to other chunks.
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64>;
 }
 
 impl RawPile for Box<dyn RawPile> {
@@ -182,6 +187,9 @@ impl RawPile for Box<dyn RawPile> {
     fn delete(&self, kind: Keyspace, key: &[u8]) -> anyhow::Result<()> {
         self.as_ref().delete(kind, key)
     }
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        self.as_ref().delete_many(kind, keys)
+    }
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -210,6 +218,10 @@ impl RawPile for Box<dyn RawPile> {
     fn describe_pipeline(&self) -> anyhow::Result<Vec<PipelineDescription>> {
         self.as_ref().describe_pipeline()
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.as_ref().chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }
 
 impl<RP: RawPile> RawPile for Arc<RP> {
@@ -225,6 +237,9 @@ impl<RP: RawPile> RawPile for Arc<RP> {
     fn delete(&self, kind: Keyspace, key: &[u8]) -> anyhow::Result<()> {
         self.as_ref().delete(kind, key)
     }
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        self.as_ref().delete_many(kind, keys)
+    }
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -253,6 +268,10 @@ impl<RP: RawPile> RawPile for Arc<RP> {
     fn describe_pipeline(&self) -> anyhow::Result<Vec<PipelineDescription>> {
         self.as_ref().describe_pipeline()
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.as_ref().chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }
 
 #[derive(Debug)]

yama/src/pile/access_guard.rs

@@ -83,6 +83,10 @@ impl<R: Clone + RawPile> RawPile for PileGuard<R> {
         bail!("Access denied");
     }
 
+    fn delete_many(&self, _kind: Keyspace, _keys: &[&[u8]]) -> anyhow::Result<()> {
+        bail!("Access denied");
+    }
+
     fn list_keys(
         &self,
         _kind: Keyspace,
@@ -130,4 +134,8 @@ impl<R: Clone + RawPile> RawPile for PileGuard<R> {
         // TODO(question) Should we be described in the pipeline?
         self.underlying.describe_pipeline()
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.underlying.chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }

yama/src/pile/compression.rs

@@ -278,6 +278,10 @@ impl<R: RawPile> RawPile for RawPileCompressor<R> {
         self.underlying.delete(kind, key)
     }
 
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        self.underlying.delete_many(kind, keys)
+    }
+
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -348,4 +352,8 @@ impl<R: RawPile> RawPile for RawPileCompressor<R> {
         });
         Ok(underlying)
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.underlying.chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }

yama/src/pile/encryption.rs

@@ -101,6 +101,10 @@ impl<R: RawPile> RawPile for RawPileEncryptor<R> {
         self.underlying.delete(kind, key)
     }
 
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        self.underlying.delete_many(kind, keys)
+    }
+
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -127,4 +131,8 @@ impl<R: RawPile> RawPile for RawPileEncryptor<R> {
         underlying.push(PipelineDescription::Encryption);
         Ok(underlying)
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.underlying.chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }

yama/src/pile/integrity.rs

@@ -98,6 +98,10 @@ impl<RP: RawPile> RawPile for RawPileIntegrityChecker<RP> {
         self.underlying.delete(kind, key)
     }
 
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        self.underlying.delete_many(kind, keys)
+    }
+
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -149,4 +153,8 @@ impl<RP: RawPile> RawPile for RawPileIntegrityChecker<RP> {
         underlying.push(PipelineDescription::Integrity);
         Ok(underlying)
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        self.underlying.chunk_id_transfer_ordering_hint(chunk_id)
+    }
 }

yama/src/pile/local_sqlitebloblogs.rs

@@ -15,20 +15,23 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
-use std::collections::hash_map::Entry;
-use std::collections::{HashMap, VecDeque};
+use std::collections::{BTreeMap, BTreeSet, HashMap, VecDeque};
 use std::convert::{TryFrom, TryInto};
-use std::fs::{read_dir, File, OpenOptions};
+use std::fs::{read_dir, remove_file, File, OpenOptions};
 use std::io::{Read, Seek, SeekFrom, Write};
+use std::os::unix::fs::MetadataExt;
 use std::path::{Path, PathBuf};
 use std::sync::{Arc, Condvar, Mutex};
+use std::time::Duration;
 use std::{fs, thread};
 
-use anyhow::{bail, Context};
+use anyhow::{bail, ensure, Context};
 use byteorder::{BigEndian, ReadBytesExt, WriteBytesExt};
+use crossbeam_channel::{Receiver, Sender};
 use log::{info, warn};
 use nix::unistd::sync;
-use rusqlite::{params, Error, ErrorCode};
+use rusqlite::ffi::ErrorCode::ConstraintViolation;
+use rusqlite::{params, Error, ErrorCode, Transaction, TransactionBehavior, NO_PARAMS};
 use rusqlite::{Connection, OptionalExtension};
 
 use crate::definitions::ChunkId;
@@ -36,10 +39,8 @@ use crate::pile::{
     ControllerMessage, DebugStatistics, Keyspace, PipelineDescription, RawPile,
     StoragePipelineSettings,
 };
-use crate::utils::bytes_to_hexstring;
-use crossbeam_channel::{Receiver, Sender};
-use rusqlite::ffi::ErrorCode::ConstraintViolation;
-use std::time::Duration;
+use crate::progress::ProgressTracker;
+use crate::utils::{bytes_to_hexstring, LruMap};
 
 /// Bloblogs will not be reused if they are already 2 GiB large.
 pub const MAX_BLOBLOG_REUSE_SIZE: u64 = 2 * 1024 * 1024 * 1024;
@@ -47,6 +48,14 @@ pub const MAX_BLOBLOG_REUSE_SIZE: u64 = 2 * 1024 * 1024 * 1024;
 /// This many pointers will be batched up for writing.
 pub const POINTER_WRITE_BATCHES: usize = 2048;
 
+/// This many bloblogs will be kept open for reading, at maximum.
+pub const BLOBLOG_MAX_READING_FILE_COUNT: usize = 128;
+
+/// Size of a blob header within a bloblog.
+/// 32 byte Chunk Id
+/// 4 byte (u32) Blob size
+pub const BLOB_HEADER_SIZE: u64 = 32 + 4;
+
 /// A file storing a log of blobs.
 /// Format:
 ///     Repeated:
@@ -136,8 +145,8 @@ pub type BloblogId = u32;
 #[derive(Debug)]
 pub struct Inner {
     next_bloblog_id: BloblogId,
-    writer_bloblogs: Vec<BloblogId>,
-    open_bloblogs: HashMap<BloblogId, Arc<Mutex<Bloblog>>>, // TODO want an LRU cache with a weak hashmap...?
+    writer_bloblogs: Vec<(BloblogId, Arc<Mutex<Bloblog>>)>,
+    reader_bloblogs: LruMap<BloblogId, Arc<Mutex<Bloblog>>>,
     connection: Connection,
     writers_in_progress: u16,
     // We batch up pointer writes because sync() performance really hurts us if we do them one by
@@ -145,14 +154,13 @@ pub struct Inner {
     queued_pointer_writes: HashMap<ChunkId, BloblogPointer>,
 }
 
-impl Inner {
-    pub fn raw_put_chunk_pointer(
-        &self,
+fn raw_put_chunk_pointer_txn(
+    txn: &Transaction,
     chunk_id: &ChunkId,
     bloblog: BloblogId,
     offset_i64: i64,
-    ) -> anyhow::Result<()> {
-        match self.connection.execute(
+) -> anyhow::Result<()> {
+    match txn.execute(
         "INSERT INTO chunks (chunk_id, bloblog, offset) VALUES (?1, ?2, ?3)",
         params![&chunk_id[..], bloblog, offset_i64],
     ) {
@@ -165,27 +173,41 @@ impl Inner {
                 );
                 Ok(())
             } else {
-                    Err(Error::SqliteFailure(e, str))?;
-                    unreachable!();
+                Err(Error::SqliteFailure(e, str).into())
             }
         }
-            other => {
-                other?;
-                unreachable!();
-            }
+        Err(other) => Err(other.into()),
     }
+}
+
+impl Inner {
+    pub fn raw_put_chunk_pointer(
+        &mut self,
+        chunk_id: &ChunkId,
+        bloblog: BloblogId,
+        offset_i64: i64,
+    ) -> anyhow::Result<()> {
+        let txn = self.connection.transaction()?;
+        raw_put_chunk_pointer_txn(&txn, chunk_id, bloblog, offset_i64)?;
+        txn.commit()?;
+        Ok(())
     }
 
     pub fn flush(&mut self) -> anyhow::Result<()> {
         // Create a non-allocated hashmap to satisfy borrow checker, then swap it in and out
         let mut queued_pointer_writes = HashMap::with_capacity(0);
         std::mem::swap(&mut self.queued_pointer_writes, &mut queued_pointer_writes);
+
+        let txn = self.connection.transaction()?;
+
         for (chunk_id, pointer) in queued_pointer_writes.drain() {
             let offset_i64 =
                 i64::try_from(pointer.offset).expect("ouch! can't turn u64 into i64...");
-            self.raw_put_chunk_pointer(&chunk_id, pointer.bloblog, offset_i64)?;
+            raw_put_chunk_pointer_txn(&txn, &chunk_id, pointer.bloblog, offset_i64)?;
         }
         std::mem::swap(&mut self.queued_pointer_writes, &mut queued_pointer_writes);
+
+        txn.commit()?;
         Ok(())
     }
 }
@@ -247,11 +269,18 @@ impl SqliteBloblogPile {
             )?;
         }
 
+        // Enable WAL mode for significantly better write performance.
+        connection.execute_batch(
+            "
+            PRAGMA journal_mode=WAL;
+        ",
+        )?;
+
         Ok(SqliteBloblogPile {
             inner: Arc::new(Mutex::new(Inner {
                 next_bloblog_id: 0,
                 writer_bloblogs: Vec::new(),
-                open_bloblogs: HashMap::new(),
+                reader_bloblogs: LruMap::new(BLOBLOG_MAX_READING_FILE_COUNT),
                 connection,
                 writers_in_progress: 0,
                 queued_pointer_writes: Default::default(),
@@ -264,23 +293,33 @@ impl SqliteBloblogPile {
 
     fn open_bloblog(&self, bloblog_id: BloblogId) -> anyhow::Result<Arc<Mutex<Bloblog>>> {
         let mut inner = self.inner.lock().unwrap();
-        Ok(match inner.open_bloblogs.entry(bloblog_id) {
-            Entry::Occupied(entry) => entry.get().clone(),
-            Entry::Vacant(entry) => {
+
+        match inner.reader_bloblogs.get(&bloblog_id) {
+            Some(bloblog) => Ok(bloblog.clone()),
+            None => {
                 let bloblog = Arc::new(Mutex::new(Bloblog::open(
                     &self.path.join(&bloblog_id.to_string()),
                 )?));
-                entry.insert(bloblog.clone());
-                bloblog
+                inner.reader_bloblogs.insert(bloblog_id, bloblog.clone());
+                Ok(bloblog)
+            }
         }
-        })
     }
 
     fn get_writing_bloblog(&self) -> anyhow::Result<(BloblogId, Arc<Mutex<Bloblog>>)> {
         let mut inner = self.inner.lock().unwrap();
-        let writing_bloblog_id: BloblogId = match inner.writer_bloblogs.pop() {
-            None => {
-                loop {
+
+        inner.writers_in_progress += 1;
+
+        if let Some(writing_bloblog) = inner.writer_bloblogs.pop() {
+            // We already have an open bloblog to give back.
+            return Ok(writing_bloblog);
+        }
+
+        // No open bloblogs to reuse; create a new one.
+        // It's very important to create a fresh one here; we definitely don't want to use a file
+        // that someone else is using!
+        let writing_bloblog_id = loop {
             let pre_inc = inner.next_bloblog_id;
             inner.next_bloblog_id += 1;
 
@@ -289,26 +328,18 @@ impl SqliteBloblogPile {
             if !bloblog_path.exists() {
                 break pre_inc;
             }
-                }
-            }
-            Some(id) => id,
         };
 
-        let result = Ok((
-            writing_bloblog_id,
-            match inner.open_bloblogs.entry(writing_bloblog_id) {
-                Entry::Occupied(entry) => entry.get().clone(),
-                Entry::Vacant(entry) => {
         let bloblog = Arc::new(Mutex::new(Bloblog::open(
             &self.path.join(&writing_bloblog_id.to_string()),
         )?));
-                    entry.insert(bloblog.clone());
-                    bloblog
-                }
-            },
-        ));
-        inner.writers_in_progress += 1;
-        result
+
+        // MAYBE FUTURE // Insert a weak reference so we can easily get a reader for this if desired.
+        // inner.open_bloblogs.insert(writing_bloblog_id, Arc::downgrade(&bloblog));
+        // For now, I don't think we actually care about reading a bloblog that we've written
+        // (at least not usually?)
+
+        Ok((writing_bloblog_id, bloblog))
     }
 
     /// Should be called once the bloblog has been finished writing to for the moment.
@@ -321,7 +352,7 @@ impl SqliteBloblogPile {
         let size = bloblog.lock().unwrap().filesize()?;
         let mut inner = self.inner.lock().unwrap();
         if size < MAX_BLOBLOG_REUSE_SIZE {
-            inner.writer_bloblogs.push(id);
+            inner.writer_bloblogs.push((id, bloblog));
         }
         inner.writers_in_progress -= 1;
         if inner.writers_in_progress == 0 {
@@ -347,8 +378,33 @@ impl SqliteBloblogPile {
             .optional()?)
     }
 
+    fn get_chunk_pointers(
+        &self,
+        chunk_ids: &[&[u8]],
+    ) -> anyhow::Result<Vec<Option<BloblogPointer>>> {
+        let mut inner = self.inner.lock().unwrap();
+        let txn = inner.connection.transaction()?;
+        let mut result = Vec::with_capacity(chunk_ids.len());
+        {
+            let mut stmt = txn.prepare("SELECT bloblog, offset FROM chunks WHERE chunk_id = ?1")?;
+            for &chunk_id in chunk_ids {
+                let bloglog_pointer: Option<BloblogPointer> = stmt
+                    .query_row(params![chunk_id], |row| {
+                        Ok(BloblogPointer {
+                            bloblog: row.get(0)?,
+                            offset: row.get::<_, i64>(1)? as u64,
+                        })
+                    })
+                    .optional()?;
+                result.push(bloglog_pointer);
+            }
+        }
+        txn.commit()?;
+        Ok(result)
+    }
+
     fn put_chunk_pointer(&self, chunk_id: &ChunkId, pointer: BloblogPointer) -> anyhow::Result<()> {
-        let inner = self.inner.lock().unwrap();
+        let mut inner = self.inner.lock().unwrap();
         let offset_i64 = i64::try_from(pointer.offset).expect("ouch! can't turn u64 into i64...");
         inner.raw_put_chunk_pointer(chunk_id, pointer.bloblog, offset_i64)
     }
@@ -472,6 +528,341 @@ impl SqliteBloblogPile {
         assert!(pointers_buffered.is_empty());
         Ok(())
     }
+
+    /// Look at the bloblogs in this pile and see where space may be reclaimable if we were to
+    /// compact.
+    ///
+    /// Next step: plan_compaction
+    pub fn analyse_for_compaction(&self) -> anyhow::Result<BTreeMap<BloblogId, BloblogStats>> {
+        let mut inner = self.inner.lock().unwrap();
+        // Lock the database right away.
+        let txn = inner
+            .connection
+            .transaction_with_behavior(TransactionBehavior::Exclusive)?;
+        let mut stmt = txn.prepare(
+            "
+            SELECT bloblog, COUNT(c.offset), COUNT(d.offset), SUM(COALESCE(d.size, 0))
+            FROM chunks c LEFT JOIN deleted d USING (bloblog, offset)
+            GROUP BY bloblog
+        ",
+        )?;
+
+        struct UnpopulatedBloblogStats {
+            pub bloblog_id: BloblogId,
+            pub chunks_total: u64,
+            pub chunks_deleted: u64,
+            pub bytes_deleted: u64,
+        }
+
+        let unpopul_bloblog_stats = stmt.query_map(NO_PARAMS, |row| {
+            Ok(UnpopulatedBloblogStats {
+                bloblog_id: row.get(0)?,
+                chunks_total: row.get::<_, i64>(1)?.try_into().expect("i64 -> u64"),
+                chunks_deleted: row.get::<_, i64>(2)?.try_into().expect("i64 -> u64"),
+                bytes_deleted: row.get::<_, i64>(3)?.try_into().expect("i64 -> u64"),
+            })
+        })?;
+
+        let mut final_stats = BTreeMap::new();
+
+        for unpopul_stat in unpopul_bloblog_stats {
+            let UnpopulatedBloblogStats {
+                bloblog_id,
+                chunks_total,
+                chunks_deleted,
+                bytes_deleted,
+            } = unpopul_stat?;
+            let bloblog_path = self.path.join(&bloblog_id.to_string());
+            let bytes_total = std::fs::metadata(&bloblog_path)
+                .with_context(|| format!("Failed to get metadata for bloblog: {:?}", bloblog_path))?
+                .size();
+
+            final_stats.insert(
+                bloblog_id,
+                BloblogStats {
+                    chunks_total,
+                    chunks_deleted,
+                    bytes_total,
+                    // Add a slight correction since we can count the blob headers of deleted blobs
+                    // as deleted.
+                    bytes_deleted: bytes_deleted + chunks_deleted * BLOB_HEADER_SIZE,
+                },
+            );
+        }
+
+        Ok(final_stats)
+    }
+
+    /// Look at the analysis of compaction and, using the specified thresholds, come up with a plan
+    /// to perform compaction.
+    ///
+    /// May return an empty plan if compaction isn't worthwhile.
+    ///
+    /// Previous step: analyse_for_compaction
+    /// Next step: perform_compaction
+    pub fn plan_compaction(
+        &self,
+        thresholds: &CompactionThresholds,
+        analysis: BTreeMap<BloblogId, BloblogStats>,
+    ) -> anyhow::Result<CompactionPlan> {
+        let bloblogs_to_replace: BTreeMap<BloblogId, BloblogStats> = analysis
+            .into_iter()
+            .filter(|(_id, stats)| thresholds.should_replace_bloblog(stats))
+            .collect();
+        let reclaimable_space: u64 = bloblogs_to_replace
+            .values()
+            .map(|bs| bs.bytes_deleted)
+            .sum();
+        let bytes_to_write: u64 = bloblogs_to_replace
+            .values()
+            .map(|bs| bs.bytes_total - bs.bytes_deleted)
+            .sum();
+        let small_bloblogs: u32 = bloblogs_to_replace
+            .values()
+            .filter(|bs| bs.bytes_total - bs.bytes_deleted < thresholds.cond_if_less_allocated_than)
+            .count() as u32;
+
+        if reclaimable_space < thresholds.minimum_to_reclaim
+            && small_bloblogs < thresholds.minimum_small_bloblogs_to_merge
+        {
+            // Nothing worth doing: return an empty plan.
+            return Ok(CompactionPlan {
+                bloblogs_to_replace: Default::default(),
+                bytes_to_write: 0,
+                reclaimable_space: 0,
+                small_bloblogs: 0,
+            });
+        }
+
+        Ok(CompactionPlan {
+            bloblogs_to_replace: bloblogs_to_replace.keys().copied().collect(),
+            bytes_to_write,
+            reclaimable_space,
+            small_bloblogs,
+        })
+    }
+
+    /// Given a compaction plan, perform the compaction.
+    /// There shouldn't be any decisions left to be made at this point: just action.
+    ///
+    /// TODO flock the bloblogs to be removed and make readers and writers also flock them too.
+    ///
+    /// TODO find a way to deal with bloblogs that are entirely unreferenced from the index
+    ///      (e.g. bloblogs that weren't written properly, e.g. if compaction fails.)
+    pub fn perform_compaction(
+        &self,
+        mut progress: Box<dyn ProgressTracker>,
+        plan: CompactionPlan,
+    ) -> anyhow::Result<()> {
+        #[derive(Copy, Clone, Ord, PartialOrd, Eq, PartialEq)]
+        struct ReplacedBlobRow {
+            pub old_bloblog: BloblogId,
+            pub old_offset: u64,
+            pub chunk_id: ChunkId,
+        }
+
+        if plan.bloblogs_to_replace.is_empty() {
+            info!("No compaction to be done.");
+            return Ok(());
+        }
+
+        let mut to_preserve = BTreeSet::new();
+        let mut replacements = BTreeMap::new();
+
+        progress.set_max_size(plan.bytes_to_write);
+
+        // First find all the blobs we need to replace.
+        {
+            let mut inner = self.inner.lock().unwrap();
+            // Lock the database right away.
+            let txn = inner
+                .connection
+                .transaction_with_behavior(TransactionBehavior::Exclusive)?;
+            let mut stmt = txn.prepare(
+                "
+                SELECT chunk_id, c.offset
+                FROM chunks c LEFT JOIN deleted d USING (bloblog, offset)
+                WHERE bloblog = ?1 AND d.offset IS NULL
+            ",
+            )?;
+            for bloblog in plan.bloblogs_to_replace.iter().copied() {
+                to_preserve.extend(
+                    stmt.query_map([bloblog], |row| {
+                        let mut chunk_id = ChunkId::default();
+                        chunk_id.copy_from_slice(row.get::<_, Vec<u8>>(0).unwrap().as_slice());
+                        Ok(ReplacedBlobRow {
+                            old_bloblog: bloblog,
+                            chunk_id,
+                            old_offset: row.get::<_, i64>(1).unwrap().try_into().unwrap(),
+                        })
+                    })?
+                    .collect::<Result<Vec<ReplacedBlobRow>, _>>()?,
+                );
+            }
+        }
+
+        // Then make the replacements
+        info!("Rewriting bloblogs...");
+        let mut buf = Vec::new();
+        let mut iterator = to_preserve.into_iter();
+        loop {
+            let (new_bloblog_id, bloglog_mutex) = self.get_writing_bloblog()?;
+            let mut new_bloblog = bloglog_mutex.lock().expect("Failed to lock bloblog?");
+            let mut is_more = false;
+
+            while let Some(preserve) = iterator.next() {
+                is_more = true;
+
+                // Get hold of the old bloblog
+                let old_bloblog = self.open_bloblog(preserve.old_bloblog)?;
+                let mut old_bloblog = old_bloblog.lock().unwrap();
+
+                // Transfer the blob
+                buf.clear();
+                old_bloblog.read_blob(preserve.old_offset, &preserve.chunk_id, &mut buf)?;
+                let new_offset = new_bloblog.write_blob(&preserve.chunk_id, &buf)?;
+
+                // Make a note of the replacement
+                replacements.insert(
+                    preserve,
+                    BloblogPointer {
+                        bloblog: new_bloblog_id,
+                        offset: new_offset,
+                    },
+                );
+
+                progress.inc_progress(buf.len() as u64);
+
+                if new_bloblog.filesize()? > MAX_BLOBLOG_REUSE_SIZE {
+                    // get a new bloblog to write with.
+                    break;
+                }
+            }
+
+            drop(new_bloblog);
+            self.return_writing_bloblog(new_bloblog_id, bloglog_mutex)?;
+
+            if !is_more {
+                break;
+            }
+        }
+
+        info!("Applying replacements...");
+        {
+            let mut inner = self.inner.lock().unwrap();
+            // Lock the database right away.
+            let txn = inner
+                .connection
+                .transaction_with_behavior(TransactionBehavior::Exclusive)?;
+            let mut stmt = txn.prepare(
+                "
+                UPDATE chunks
+                SET bloblog = ?1, offset = ?2
+                WHERE chunk_id = ?3
+            ",
+            )?;
+
+            for (replacement_row, new_pos) in replacements {
+                ensure!(
+                    stmt.execute(params![
+                        new_pos.bloblog,
+                        new_pos.offset as i64,
+                        &replacement_row.chunk_id as &[u8]
+                    ])? == 1,
+                    "Wrong number of rows updated for replacement!"
+                );
+            }
+
+            drop(stmt);
+            txn.commit().context("committing replacements")?;
+        }
+
+        // TODO fsync new bloblogs
+
+        info!("Deleting old bloblogs...");
+        {
+            let mut inner = self.inner.lock().unwrap();
+            // Lock the database right away.
+            let txn = inner
+                .connection
+                .transaction_with_behavior(TransactionBehavior::Exclusive)?;
+
+            for bloblog_id in plan.bloblogs_to_replace.iter().copied() {
+                let deleted_chunks = txn.execute(
+                    "
+                    DELETE FROM chunks WHERE bloblog = ?1
+                ",
+                    params![bloblog_id],
+                )?;
+
+                let deleted_deleted = txn.execute(
+                    "
+                    DELETE FROM deleted WHERE bloblog = ?1
+                ",
+                    params![bloblog_id],
+                )?;
+
+                ensure!(deleted_chunks == deleted_deleted, "Undeleted chunks left in bloblog {bloblog_id}: CHUNKS={deleted_chunks} DELETED={deleted_deleted}");
+
+                let bloblog_path = self.path.join(bloblog_id.to_string());
+                remove_file(&bloblog_path).with_context(|| {
+                    format!("Failed to remove obsolete bloblog: {:?}", bloblog_path)
+                })?;
+            }
+
+            txn.commit()?;
+        }
+
+        Ok(())
+    }
+}
+
+pub struct BloblogStats {
+    pub chunks_total: u64,
+    pub chunks_deleted: u64,
+    pub bytes_total: u64,
+    pub bytes_deleted: u64,
+}
+
+pub struct CompactionPlan {
+    pub bloblogs_to_replace: BTreeSet<BloblogId>,
+    pub bytes_to_write: u64,
+    pub reclaimable_space: u64,
+    pub small_bloblogs: u32,
+}
+
+pub struct CompactionThresholds {
+    /// Minimum bytes to be reclaimable overall for compaction to be worthwhile.
+    pub minimum_to_reclaim: u64,
+
+    /// (alternative reason) Minimum number of files to be undersized in order for compaction
+    /// to be worthwhile.
+    /// This gives us a way to make compaction run if we have lots of tiny bloblogs.
+    pub minimum_small_bloblogs_to_merge: u32,
+
+    /// A bloblog will be replaced if the deallocated size is greater than this.
+    pub cond_if_more_deallocated_than: u64,
+
+    /// A bloblog will be replaced if the allocated size is less than this.
+    pub cond_if_less_allocated_than: u64,
+}
+
+impl CompactionThresholds {
+    pub fn should_replace_bloblog(&self, bloblog_stats: &BloblogStats) -> bool {
+        let allocated = bloblog_stats.bytes_total - bloblog_stats.bytes_deleted;
+        // Note that this will also trigger for fully-deallocated files if
+        let is_small = allocated < self.cond_if_less_allocated_than;
+        let has_large_deallocations =
+            bloblog_stats.bytes_deleted > self.cond_if_more_deallocated_than;
+        is_small || has_large_deallocations
+    }
+}
+
+pub struct CompactionOutcome {
+    pub bloblogs_deleted: u32,
+    pub bloblogs_created: u32,
+    pub bytes_deleted: u32,
+    pub bytes_created: u32,
 }
 
 impl Drop for SqliteBloblogPile {
@@ -616,6 +1007,59 @@ impl RawPile for SqliteBloblogPile {
             }
         }
     }
+
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        match kind {
+            Keyspace::Chunk => {
+                let mut chunk_pointers_by_bloblog: BTreeMap<BloblogId, Vec<(u64, &[u8])>> =
+                    BTreeMap::new();
+
+                for (chunk_pointer, chunk_id) in self
+                    .get_chunk_pointers(keys)
+                    .context("failed to get chunk pointers")?
+                    .into_iter()
+                    .zip(keys)
+                    .filter_map(|(pointer, &chunk_id)| match pointer {
+                        Some(pointer) => Some((pointer, chunk_id)),
+                        None => None,
+                    })
+                {
+                    chunk_pointers_by_bloblog
+                        .entry(chunk_pointer.bloblog)
+                        .or_default()
+                        .push((chunk_pointer.offset, chunk_id));
+                }
+                let mut inner = self.inner.lock().unwrap();
+                let txn = inner.connection.transaction()?;
+                {
+                    let mut stmt = txn.prepare(
+                        "INSERT OR IGNORE INTO deleted (bloblog, offset, size)
+                            VALUES (?1, ?2, ?3)",
+                    )?;
+                    for (bloblog_id, entries) in chunk_pointers_by_bloblog {
+                        let bloblog_mutex = self.open_bloblog(bloblog_id)?;
+                        let mut bloblog = bloblog_mutex.lock().unwrap();
+                        for (chunk_offset, raw_chunk_id) in entries {
+                            let mut chunk_id: ChunkId = Default::default();
+                            chunk_id.copy_from_slice(raw_chunk_id);
+                            let size = bloblog.blob_len(chunk_offset, &chunk_id)?;
+                            let offset_i64 = i64::try_from(chunk_offset)
+                                .expect("ouch! can't turn u64 into i64...");
+                            stmt.execute(params![bloblog_id, offset_i64, size])?;
+                        }
+                    }
+                }
+                txn.commit().context("Failed to commit chunk deletions")?;
+            }
+            _ => {
+                for &key in keys {
+                    self.delete(kind, key)?;
+                }
+            }
+        }
+        Ok(())
+    }
+
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -730,6 +1174,16 @@ impl RawPile for SqliteBloblogPile {
     fn describe_pipeline(&self) -> anyhow::Result<Vec<PipelineDescription>> {
         Ok(vec![PipelineDescription::Store])
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        let chunk_pointer = self
+            .get_chunk_pointer(chunk_id)?
+            .context("Can't get chunk ID transfer ordering hint for chunk without pointer.")?;
+
+        // Scheme: 24-bit bloblog ID
+        // followed by 40-bit offset
+        Ok(((chunk_pointer.bloblog as u64) << 40) | (chunk_pointer.offset & 0xFF_FF_FF_FF_FF))
+    }
 }
 
 struct KeyIterator {
@@ -775,9 +1229,10 @@ impl Iterator for KeyIterator {
 
 #[cfg(test)]
 mod tests {
-    use crate::pile::local_sqlitebloblogs::Bloblog;
     use temp_dir::TempDir;
 
+    use crate::pile::local_sqlitebloblogs::Bloblog;
+
     #[test]
     pub fn bloblog_read_write_test() {
         let td = TempDir::new().unwrap();

yama/src/remote/requester.rs

@@ -4,7 +4,7 @@ use std::sync::{Arc, Mutex};
 use std::thread;
 use std::thread::JoinHandle;
 
-use anyhow::anyhow;
+use anyhow::{anyhow, bail};
 use crossbeam_channel::{Receiver, Sender};
 use log::{error, info};
 
@@ -307,6 +307,12 @@ impl RawPile for Requester {
             other => Err(anyhow!("Received {:?} for Delete", other)),
         }
     }
+    fn delete_many(&self, kind: Keyspace, keys: &[&[u8]]) -> anyhow::Result<()> {
+        for &key in keys {
+            self.delete(kind, key)?;
+        }
+        Ok(())
+    }
     fn list_keys(
         &self,
         kind: Keyspace,
@@ -438,6 +444,10 @@ impl RawPile for Requester {
             other => Err(anyhow!("Received {:?} for Describe", other)),
         }
     }
+
+    fn chunk_id_transfer_ordering_hint(&self, _chunk_id: &ChunkId) -> anyhow::Result<u64> {
+        bail!("You probably shouldn't be using chunk ID transfer ordering hints with a remote.");
+    }
 }
 
 pub struct ListKeyIterator {

yama/src/tree.rs

@@ -185,7 +185,7 @@ pub fn differentiate_node_in_place(new: &mut TreeNode, old: &TreeNode) -> anyhow
 /// result is in-place.
 ///
 /// Preconditions:
-/// - `old` must be an integrated pointer.
+/// - `old` must be an integrated pointer. (Otherwise this algorithm is not correct.)
 /// - `old` is the parent of `new`
 pub fn integrate_node_in_place(new: &mut TreeNode, old: &TreeNode) -> anyhow::Result<()> {
     if let TreeNode::Directory { children, .. } = new {

yama/src/utils.rs

@@ -15,6 +15,7 @@ You should have received a copy of the GNU General Public License
 along with Yama.  If not, see <https://www.gnu.org/licenses/>.
 */
 
+use std::collections::{BTreeMap, BTreeSet};
 use std::fmt::Write;
 
 pub fn bytes_to_hexstring(chunkid: &[u8]) -> String {
@@ -42,3 +43,98 @@ pub fn get_number_of_workers(first_try_env_name: &str) -> u8 {
         }
     }
 }
+
+#[derive(Clone, Debug)]
+pub struct LruMap<K, V> {
+    capacity: usize,
+    last_access: BTreeSet<(u64, K)>,
+    items: BTreeMap<K, (V, u64)>,
+    counter: u64,
+}
+
+impl<K: Ord + Clone, V> LruMap<K, V> {
+    pub fn new(capacity: usize) -> LruMap<K, V> {
+        LruMap {
+            capacity,
+            last_access: BTreeSet::new(),
+            items: BTreeMap::new(),
+            counter: 0,
+        }
+    }
+
+    /// Gets an item from the LRU map.
+    pub fn get(&mut self, key: &K) -> Option<&V> {
+        match self.items.get_mut(key) {
+            Some((value, last_used_instant)) => {
+                assert!(
+                    self.last_access.remove(&(*last_used_instant, key.clone())),
+                    "Corrupt LRU map: freshen not correct."
+                );
+                let new_instant = self.counter;
+                self.counter += 1;
+                self.last_access.insert((new_instant, key.clone()));
+                *last_used_instant = new_instant;
+                Some(value)
+            }
+            None => None,
+        }
+    }
+
+    pub fn insert(&mut self, key: K, value: V) -> Option<V> {
+        let new_instant = self.counter;
+        self.counter += 1;
+
+        let retval = match self.items.insert(key.clone(), (value, new_instant)) {
+            Some((old_entry, old_instant)) => {
+                assert!(
+                    self.last_access.remove(&(old_instant, key.clone())),
+                    "Corrupt LRU map: insert not correct."
+                );
+                Some(old_entry)
+            }
+            None => None,
+        };
+        self.last_access.insert((new_instant, key));
+
+        if retval.is_none() {
+            // We didn't replace any item, so we have grown by 1.
+            // Check if we need to evict.
+            if self.items.len() > self.capacity {
+                self.evict();
+            }
+        }
+
+        retval
+    }
+
+    pub fn evict(&mut self) -> Option<(K, V)> {
+        if let Some(first_entry) = self.last_access.iter().next().cloned() {
+            self.last_access.remove(&first_entry);
+            let (_, key) = first_entry;
+            let (value, _) = self
+                .items
+                .remove(&key)
+                .expect("Corrupt LRU map: last access and items out of sync");
+
+            Some((key, value))
+        } else {
+            None
+        }
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use crate::utils::LruMap;
+
+    #[test]
+    fn test_lru_map() {
+        let mut lmap = LruMap::new(3);
+        lmap.insert(1, 1);
+        lmap.insert(2, 1);
+        lmap.insert(3, 1);
+        assert_eq!(lmap.get(&1), Some(&1));
+        lmap.insert(4, 1);
+        assert_eq!(lmap.get(&2), None);
+    }
+}