microcosm.blue
Fast and robust atproto CAR file processing in rust
reduce fjall mem
authored by
+9
-14
+3
-1
examples/disk-read-file/main.rs
+3
-1
examples/disk-read-file/main.rs
···
5
5
extern crate repo_stream;
6
6
7
7
use mimalloc::MiMalloc;
8
-
9
8
#[global_allocator]
10
9
static GLOBAL: MiMalloc = MiMalloc;
11
10
···
58
57
// at this point you might want to fetch the account's signing key
59
58
// via the DID from the commit, and then verify the signature.
60
59
log::warn!("big's comit ({:?}): {:?}", t0.elapsed(), commit);
60
+
61
+
// log::info!("now is good time to check mem usage...");
62
+
// tokio::time::sleep(std::time::Duration::from_secs(15)).await;
61
63
62
64
// pop the driver back out to get some code indentation relief
63
65
driver
+5
-12
src/disk.rs
+5
-12
src/disk.rs
···
19
19
20
20
use crate::Bytes;
21
21
use crate::drive::DriveError;
22
-
use fjall::config::{CompressionPolicy, PinningPolicy, RestartIntervalPolicy};
23
-
use fjall::{CompressionType, Database, Error as FjallError, Keyspace, KeyspaceCreateOptions};
22
+
use fjall::{Database, Error as FjallError, Keyspace, KeyspaceCreateOptions};
24
23
use std::path::PathBuf;
25
24
26
25
#[derive(Debug, thiserror::Error)]
···
111
110
let max_stored = max_stored_mb * 2_usize.pow(20);
112
111
let (db, partition) = tokio::task::spawn_blocking(move || {
113
112
let db = Database::builder(path)
114
-
// .manual_journal_persist(true)
115
-
// .flush_workers(1)
116
-
// .compaction_workers(1)
117
-
.journal_compression(CompressionType::None)
118
-
.cache_size(cache_mb as u64 * 2_u64.pow(20))
113
+
.manual_journal_persist(true)
114
+
.worker_threads(1)
115
+
.cache_size(cache_mb as u64 * 2_u64.pow(20) / 2)
119
116
.temporary(true)
120
117
.open()?;
121
118
let opts = KeyspaceCreateOptions::default()
122
-
.data_block_restart_interval_policy(RestartIntervalPolicy::all(8))
123
-
.filter_block_pinning_policy(PinningPolicy::disabled())
124
119
.expect_point_read_hits(true)
125
-
.data_block_compression_policy(CompressionPolicy::disabled())
126
-
.manual_journal_persist(true)
127
-
.max_memtable_size(32 * 2_u64.pow(20));
120
+
.max_memtable_size(16 * 2_u64.pow(20));
128
121
let partition = db.keyspace("z", || opts)?;
129
122
130
123
Ok::<_, DiskError>((db, partition))
+1
-1
src/drive.rs
+1
-1
src/drive.rs
···
376
376
let maybe_processed = MaybeProcessedBlock::maybe(self.process, data);
377
377
mem_size += maybe_processed.len();
378
378
chunk.push((cid, maybe_processed));
379
-
if mem_size >= self.max_size {
379
+
if mem_size >= (self.max_size / 2) {
380
380
// soooooo if we're setting the db cache to max_size and then letting
381
381
// multiple chunks in the queue that are >= max_size, then at any time
382
382
// we might be using some multiple of max_size?