* filer/leveldb_handle: reopen a database whose journal latched a write error
goleveldb keeps the first journal write error and returns it for every
later write until the database is reopened, so one transient ENOSPC or
EIO fails all writes until a restart. The handle offers the leveldb.DB
methods the stores use; a write that fails on an OS error probes the
storage and, once it accepts writes and no call holds the database,
closes and reopens it and retries the write once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer: open the leveldb stores through leveldb_handle
leveldb, leveldb2 and leveldb3 now recover from a transient storage
write error without a restart.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer: leveldb stores return iterator errors from listings and deletes
A listing on a database that cannot be read reported an empty
directory, and DeleteFolderChildren wrote an empty batch, because the
iterator error was never checked.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer/leveldb_handle: test a reopen that fails after the probe passes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer/leveldb_handle: probe storage before draining readers
A probe that stalls in Create, Write or Sync held mu exclusively and
blocked every read on the handle. Reopen attempts are now serialized by
their own mutex, and the probe runs before the drain.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer/leveldb_handle: hold new calls back while a reopen drains
TryLock succeeds only at an instant with no reader, so overlapping
listings could keep a reopen from ever draining while each write spun
for the full drain timeout. The drain now holds new calls back for at
most drainTimeout; a call nested inside a running one waits that out
instead of deadlocking. A drain that times out holds off the next
attempt for busyReopenInterval, so new calls are not held back over and
over.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer/leveldb_handle: bound the wait for a running reopen
A write that hit the latched error waited on the reopen mutex for as long
as another caller's storage probe ran, so a stalled probe queued every
writer. Callers now wait at most reopenWait and return their error, and
still retry once a concurrent reopen succeeds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* filer/leveldb_handle: tests wait for the drain and readers explicitly
The nested-read test polled for a drain that could start and end between
polls and then spun forever; it now syncs on the probe, widens the drain
and fails if the put returns first. The overlapping-readers test waits
until every reader holds an iterator before it writes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* test: prefixed directory listing with start below the prefix
A listing constrained to prefix "item-b" and resumed from "item" must
return item-b.versions. The seek-based stores return an empty page
because the iterator lands before the prefix range and breaks on the
first non-matching name.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* filer: clamp prefixed-listing seek to the prefix range
Seek-based stores (leveldb, leveldb2, leveldb3, rocksdb, tikv, etcd)
started the scan at startFileName and stopped at the first key outside
the prefix. A startFileName below the prefix landed the iterator on a
non-matching sibling first, so the listing returned an empty page even
though matching entries followed. etcd had the mirror problem: a range
beginning below the prefix end leaked non-matching entries. Begin the
seek at max(startFileName, prefix), which is what foundationdb and the
sql stores already do.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* filer(hbase): honor startFileName lower bound in prefixed listings
The scan starts at the prefix but only skipped the exact startFileName,
so entries sorting before it were returned again on resume.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* filer: keep paging a prefixed listing while matches can still follow
The fallback for stores without native prefix support stopped paging as
soon as the last scanned name reached the prefix, dropping prefix
matches that continue on the next page.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* filer: return the last delivered name from the prefixed-listing fallback
The fallback for stores without native prefix support pages raw entries
and filters client-side. Keep the scan cursor and the returned cursor
separate: resuming must continue after the last delivered match, not the
last scanned name, or matches still inside the current raw page are
skipped on the next page. Also keep paging while the scan cursor is
inside or below the prefix range instead of stopping at the first name
that reaches the prefix.
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* filer(hbase): start prefixed-listing scans at the resume key
Generated with [Devin](https://devin.ai)
Co-Authored-By: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
---------
Co-authored-by: Devin <158243242+devin-ai-integration[bot]@users.noreply.github.com>
* volume: never compact a volume tiered to remote storage
A tiered volume's data backend is the remote object, but compaction reads
the local .dat path. When tiering removed the local .dat, every vacuum of a
tiered volume with garbage fails with "no such file or directory", and the
master's sweep and the worker's vacuum task keep selecting it. When the
local .dat was kept, the compaction succeeds against the stale copy and the
commit reloads the volume to read the remote object (old layout) through the
compacted index, so live needles stop reading.
Mirror the Rust volume server (#11545):
- CompactByIndex and CompactByVolumeData refuse a volume with a remote file
before creating .cpd/.cpx ("volume N is tiered to remote storage, cannot
compact").
- CommitCompact refuses under the data-file locks when the volume became
tiered during the copy, and removes .cpd/.cpx.
- Store.CheckCompactVolume reports no garbage for a tiered volume, so the
master's vacuum and volume.vacuum do not select it.
- The worker's vacuum detection skips a volume with a remote copy, as balance
and EC detection already do.
Tests (each fails on master): TestCompactRefusesTieredVolume,
TestCommitCompactRefusesVolumeTieredDuringCopy (on master the commit
succeeds and a surviving needle no longer reads),
TestPickPrimaryReplicaSkipsTieredVolumes.
* volume: skip tiered replicas in vacuum selection, keep their real ratio
Reporting a zero garbage ratio for a tiered volume hid it from
volume.tier.compact, whose preflight reads that ratio while the volume is
still remote, and did not exclude it from a vacuum with a zero threshold.
Report the real ratio again, as the Rust volume server does, and have the
master's vacuum check skip a replica whose volume info names remote
storage, whatever the threshold or request.
When CommitCompact refuses a volume tiered during the copy, also close the
temporary needle map and remove .cpldb: nothing else would, since no later
vacuum selects the volume.
The tiered-compaction test no longer depends on random needle sizes
reaching a 0.3 garbage ratio.
* volume tier: do not attach a copy of a .dat a compaction replaced
The upload copies the .dat, then records the remote file in the .vif and
switches the backend. A compaction commit in between swaps in a compacted
.dat and index; the upload then served the old-layout remote object through
the new index, so live needles stopped reading. CommitCompact's tiered
check cannot see this: the volume is not tiered until the switch.
AttachRemoteDatFile records the remote file and switches the backend under
the data-file locks, and refuses when the backend is no longer the file
that was copied. The upload then removes the unused remote copy.
* volume tier: keep a committed .vif entry when only the dir fsync fails
AttachRemoteDatFile rolled back the in-memory remote entry on any
SaveVolumeInfo error. A NotCrashDurableError comes after the rename, so
the .vif on disk already names the copy; with the entry gone, the upload
handler deleted that object, and a reload would serve from a missing
remote file. Keep the entry and switch to the copy as the .vif says,
and still return the error so the caller keeps the local .dat.
* vacuum worker: filter tiered replicas instead of skipping the volume
A partially tiered volume is the leftover of an interrupted
volume.tier.upload; skipping the whole volume left garbage on its local
replicas unreclaimed. Per-replica filtering matches the master's
batchVacuumVolumeCheck and is safe: a tiered replica refuses both
compaction and commit.
* filer: sync leveldb entry writes when the path rule asks for fsync
The fsync storage rule only reached the volume upload; the metadata
writes behind an acknowledged create, update or delete still went to
the leveldb journal with Sync:false, so a power loss could roll back
an acknowledged bucket or object entry.
A write-side flag on the entry context carries the request: the filer
sets it from the path's storage rule in CreateEntry, UpdateEntry,
EnsureDirectoryEntry and DeleteEntryMetaAndData, and the HTTP write
path sets it for an explicit ?fsync=true. The leveldb stores write
with Sync:true when the flag is present.
* volume: flush the .idx before acknowledging a durable write
A write with fsync synced only the .dat: the index row written on
publish stayed in the page cache, so a power loss could leave the
needle's bytes durable with nothing indexing them.
The durable paths now run one .idx sync after publish, before the
requests complete — one per syncWrite and one per batch run, paired
with the .dat sync, the same split the Rust volume server makes
between flush_dat and flush_idx. A failed .idx sync cannot be rolled
back, so it stops the volume taking writes and fails the requests
instead of acknowledging an index entry that may not survive.
---------
Co-authored-by: Javier de la Torre <jatorre@lakehousebox.com>
* filer: decode a listed entry without building its chunk list
A readdir reads attributes and never looks at chunks, but decoding an
entry builds the whole chunk list first: four allocations per chunk, all
of it thrown away. On a directory of ordinary 4MB-chunked files that is
most of what listing costs.
DecodeAttributesOnly walks the wire format and hands everything except
the chunks to the generated unmarshaller, so new fields in filer.proto
need no attention here. The chunks are still measured, because the S3
copy and multipart paths deliberately store a zero FileSize and let the
chunk extents define the size, but nothing is allocated to do it.
The blob is only re-encoded once a chunk is actually seen, so an entry
without any -- every directory, for one -- is unmarshalled where it lies
and pays nothing for the walk.
Listings opt in through the context, the way the lazy remote paths
already do; a store that ignores it stays correct.
chunks full attrs-only allocs
0 312.8n 310.1n ~ 1 -> 1
1 686.1n 411.1n -40.07% 7 -> 1
4 1.742u 667.4n -61.69% 24 -> 1
16 5.770u 1.544u -73.25% 86 -> 1
64 25.23u 6.004u -76.20% 328 -> 1
* mount: list directories with chunk lists omitted
The two meta cache listings behind a readdir are the only callers, and
neither reads a chunk. On 200k single-chunk files one enumeration goes
from 364ms to 277ms and drops a million allocations.
The read-through listing still fetches whole entries from the filer,
which would need the request to say it wants attributes only.
* mount: give the readdir benchmark's entries a chunk
Chunkless entries made the decode look far cheaper than it is, which is
the part of a listing worth measuring.
* filer: let a listing ask for entries without their chunk lists
The read-through readdir fetches whole entries over gRPC, and for a wide
directory the chunk lists are most of what crosses the wire and most of
what the client then unmarshals. A 4MB-chunked file is 113 bytes of
entry against 46 without its chunk.
ListEntriesRequest gains omit_chunks. The size a client needs is already
in the attributes, where the store decode folded the chunk extents in,
so dropping the list costs the client nothing.
The filer still reads the entries whole. A listing is where a TTL-expired
entry gets collected and deleted, and deleting one needs its chunks to
find the data, so omitting them there would leak. Only the response is
trimmed.
The hint moves to filer_pb so one context flag serves both transports:
the gRPC request sets omit_chunks, and a listing served from the local
store skips building the chunks. Cache population is unaffected either
way, since EnsureVisited starts from its own context.
* filer: reject a chunk the full decoder would reject
The walk skipped a chunk's bytes without looking inside them, so a
FileChunk carrying a corrupt nested fid, or a string that is not valid
UTF-8, sailed past the listing decoder while every other read of the same
entry still failed. The file listed with a plausible size and then gave
EIO on open, and corruption that used to fail the listing loudly was
hidden instead.
The chunk bytes are the one part of the blob the generated unmarshaller
never sees, so the two checks it would have made are made here: a
submessage has to parse, and a proto3 string has to be valid UTF-8.
FileChunk's only submessages are FileIds of scalars, so walking them is a
complete check. A descriptor-driven test fails if FileChunk ever gains a
field of either kind that the walk does not know to check, which is the
part that keeps this honest as filer.proto grows.
Taking the scratch buffer lazily, only once a chunk is actually dropped,
also takes the pool out of the path for entries that have none. Those
were measurably slower than the full decoder before; they are now level
with it. Each chunk's length prefix is parsed once rather than twice.
chunks full attrs-only vs base
0 171.4n 176.9n ~ (p=0.670)
1 366.6n 259.4n -29.24%
4 1.034u 500.2n -51.60%
16 3.905u 1.464u -62.52%
64 13.48u 5.195u -61.46%
* filer: carry the size before dropping chunks over the wire
Dropping the chunk list assumed every store folds the chunk extents into
FileSize when it decodes. A store that keeps entries as JSON rather than
as an encoded Entry never re-derives it, so an object written with a zero
FileSize kept its real size only in the chunks, and stripping them left
the client reading the file as empty. Stamp the size into the attributes
first, which costs nothing and does not depend on how the store loaded
the entry.
* mount: test that the readdir context reaches the store decode
Everything else exercises the decoder directly, so a refactor that
stopped threading the context would have reverted the whole thing with
every test still passing.
The benchmark's chunks also carried a constant legacy FileId, which
BeforeEntrySerialization reparses over Fid on the way in, so all 200k
entries stored one byte-identical chunk rather than the varying fixture
it looked like.
* mount: add singleflight to deduplicate concurrent EnsureVisited calls
When multiple goroutines access the same uncached directory simultaneously,
they would all make redundant network requests to the filer. This change
uses singleflight.Group to ensure only one goroutine fetches the directory
entries while others wait for the result.
This fixes a race condition where concurrent lookups or readdir operations
on the same uncached directory would:
1. Make duplicate network requests to the filer
2. Insert duplicate entries into LevelDB cache
3. Waste CPU and network bandwidth
* mount: fetch parent directories in parallel during EnsureVisited
Previously, when accessing a deep path like /a/b/c/d, the parent directories
were fetched serially from target to root. This change:
1. Collects all uncached directories from target to root first
2. Fetches them all in parallel using errgroup
3. Relies on singleflight (from previous commit) for deduplication
This reduces latency when accessing deep uncached paths, especially in
high-latency network environments where parallel requests can significantly
improve performance.
* mount: add batch inserts for LevelDB meta cache
When populating the meta cache from filer, entries were inserted one-by-one
into LevelDB. This change:
1. Adds BatchInsertEntries method to LevelDBStore that uses LevelDB's
native batch write API
2. Updates MetaCache to keep a direct reference to the LevelDB store
for batch operations
3. Modifies doEnsureVisited to collect entries and insert them in
batches of 100 entries
Batch writes are more efficient because:
- Reduces number of individual write operations
- Reduces disk syncs
- Improves throughput for large directories
* mount: fix potential nil dereference in MarkChildrenCached
Add missing check for inode existence in inode2path map before accessing
the InodeEntry. This prevents a potential nil pointer dereference if the
inode exists in path2inode but not in inode2path (which could happen due
to race conditions or bugs).
This follows the same pattern used in IsChildrenCached which properly
checks for existence before accessing the entry.
* mount: fix batch flush when last entry is hidden
The previous batch insert implementation relied on the isLast flag to flush
remaining entries. However, if the last entry is a hidden system entry
(like 'topics' or 'etc' in root), the callback returns early and the
remaining entries in the batch are never flushed.
Fix by:
1. Only flush when batch reaches threshold inside the callback
2. Flush any remaining entries after ReadDirAllEntries completes
3. Use error wrapping instead of logging+returning to avoid duplicate logs
4. Create new slice after flush to allow GC of flushed entries
5. Add documentation for batchInsertSize constant
This ensures all entries are properly inserted regardless of whether
the last entry is hidden, and prevents memory retention issues.
* mount: add context support for cancellation in EnsureVisited
Thread context.Context through the batch insert call chain to enable
proper cancellation and timeout support:
1. Use errgroup.WithContext() so if one fetch fails, others are cancelled
2. Add context parameter to BatchInsertEntries for consistency with InsertEntry
3. Pass context to ReadDirAllEntries for cancellation during network calls
4. Check context cancellation before starting work in doEnsureVisited
5. Use %w for error wrapping to preserve error types for inspection
This prevents unnecessary work when one directory fetch fails and makes
the batch operations consistent with the existing context-aware APIs.