Open nicolastakashi opened 7 months ago
@nicolastakashi Can you share the actual panic stacktrace? I think I saw something similar last week
We have:
bufPooled, err := r.block.chunkPool.Get(r.block.estimatedMaxChunkSize)
if err == nil {
buf = *bufPooled
} else {
buf = make([]byte, r.block.estimatedMaxChunkSize)
}
and the crash happens here few lines later
chunkLen = r.block.estimatedMaxChunkSize
if i+1 < len(pIdxs) {
if diff = pIdxs[i+1].offset - pIdx.offset; int(diff) < chunkLen {
chunkLen = int(diff)
}
}
-> cb := buf[:chunkLen]
Is the pool returning too small size slices?
I rolled back to 0.31 and seems it's not happening on this version
This is also happening to us: Thanos: v0.32.5 (deployed through bitnami helm chart) Prometheus: v2.45.1
The problem only occurs on queries which are a little bit older and only rely on the store s3 data.
panic: runtime error: slice bounds out of range [:16000] with capacity 2422
goroutine 2641055 [running]:
github.com/thanos-io/thanos/pkg/store.(*bucketChunkReader).loadChunks(0xc24d3caae0, {0x2d0fc20, 0xc338675680}, {0xc362b24000, 0x9, 0x0?}, {0xc26a0f8f80, 0x1, 0x2}, 0x1, ...)
/bitnami/blacksmith-sandox/thanos-0.32.5/src/github.com/thanos-io/thanos/pkg/store/bucket.go:3342 +0x11e5
github.com/thanos-io/thanos/pkg/store.(*bucketChunkReader).load.func4()
/bitnami/blacksmith-sandox/thanos-0.32.5/src/github.com/thanos-io/thanos/pkg/store/bucket.go:3270 +0xff
golang.org/x/sync/errgroup.(*Group).Go.func1()
/bitnami/blacksmith-sandox/thanos-0.32.5/pkg/mod/golang.org/x/sync@v0.3.0/errgroup/errgroup.go:75 +0x64
created by golang.org/x/sync/errgroup.(*Group).Go
/bitnami/blacksmith-sandox/thanos-0.32.5/pkg/mod/golang.org/x/sync@v0.3.0/errgroup/errgroup.go:72 +0xa5
I can also provide traces if that would be helpful.
how is storage gateway configured?
@MichaHoffmann This is mine.
- store
- '--log.level=warn'
- '--log.format=json'
- '--grpc-address=0.0.0.0:10901'
- '--http-address=0.0.0.0:10902'
- '--data-dir=/data'
- '--objstore.config-file=/conf/objstore.yml'
- '--index-cache.config-file=/conf/index-cache.yml'
- '--max-time=-48h'
- '--min-time=-720h'
- '--grpc-grace-period=5s'
- '--store.enable-index-header-lazy-reader'
This is our config:
- store
- '--log.level=info'
- '--log.format=logfmt'
- '--grpc-address=0.0.0.0:10901'
- '--http-address=0.0.0.0:10902'
- '--data-dir=/data'
- '--objstore.config-file=/conf/objstore.yml'
- |
--tracing.config=type: OTLP
config:
client_type: grpc
service_name: "thanos-storegateway"
endpoint: 127.0.0.1:4317
insecure: true
compression: gzip
We're also seeing this issue on one of our storagegateways sets, out of 6
Prometheus: v2.49.1 Thanos sidecar: v0.28.1 Thanos: v0.34.0
panic: runtime error: slice bounds out of range [:1027] with capacity 1024
goroutine 303830 [running]:
github.com/thanos-io/thanos/pkg/store.(*bucketChunkReader).loadChunks(0xc06bb0d980, {0x35ee848, 0xc1421085a0}, {0xc1d4e00000, 0x1849, 0x100000001?}, {0xc07006e2f8, 0x1, 0x2}, 0x2, ...)
/bitnami/blacksmith-sandox/thanos-0.34.0/src/github.com/thanos-io/thanos/pkg/store/bucket.go:3532 +0x1118
github.com/thanos-io/thanos/pkg/store.(*bucketChunkReader).load.func4()
/bitnami/blacksmith-sandox/thanos-0.34.0/src/github.com/thanos-io/thanos/pkg/store/bucket.go:3466 +0x11b
golang.org/x/sync/errgroup.(*Group).Go.func1()
/bitnami/blacksmith-sandox/thanos-0.34.0/pkg/mod/golang.org/x/sync@v0.5.0/errgroup/errgroup.go:75 +0x56
created by golang.org/x/sync/errgroup.(*Group).Go in goroutine 295848
/bitnami/blacksmith-sandox/thanos-0.34.0/pkg/mod/golang.org/x/sync@v0.5.0/errgroup/errgroup.go:72 +0x96
- name: storegateway
image: docker.io/bitnami/thanos:0.34.0-debian-11-r0
args:
- store
- '--log.level=info'
- '--log.format=logfmt'
- '--grpc-address=0.0.0.0:10901'
- '--http-address=0.0.0.0:10902'
- '--data-dir=/data'
- '--objstore.config-file=/conf/objstore.yml'
Still seeing this issue on 0.33.0
panic: runtime error: slice bounds out of range [:16000] with capacity 1249goroutine 15326510 [running]:github.com/thanos-io/thanos/pkg/store.(*bucketChunkReader).loadChunks(0x40032faae0, {0x26c6c28, 0x40248fcf00}, {0x4023fe8000, 0x94, 0x2bf1bc?}, {0x41ced43ba8, 0x1, 0x2}, 0x3997b40?, ...) /app/pkg/store/bucket.go:3504 +0xd18github.com/thanos-io/thanos/pkg/store.(*bucketChunkReader).load.func4() /app/pkg/store/bucket.go:3438 +0x84golang.org/x/sync/errgroup.(*Group).Go.func1() /go/pkg/mod/golang.org/x/sync@v0.3.0/errgroup/errgroup.go:75 +0x58created by golang.org/x/sync/errgroup.(*Group).Go in goroutine 15326102 /go/pkg/mod/golang.org/x/sync@v0.3.0/errgroup/errgroup.go:72 +0x98
So far we fixed it ourself in pkg.store, line 3532:
chunkLen = r.block.estimatedMaxChunkSize
if i+1 < len(pIdxs) {
if diff = pIdxs[i+1].offset - pIdx.offset; int(diff) < chunkLen {
chunkLen = int(diff)
}
}
// Fix: If we are about to read a chunk that is bigger than the buffer capacity,
// we need to make sure we have enough space in the buffer.
if cap(buf) < chunkLen {
// Put the current buffer back to the pool.
r.block.chunkPool.Put(&buf)
// Get a new, bigger, buffer from the pool.
bufPooled, err = r.block.chunkPool.Get(chunkLen)
if err == nil {
buf = *bufPooled
} else {
buf = make([]byte, chunkLen)
}
}
// Fix: end
cb := buf[:chunkLen]
n, err = io.ReadFull(bufReader, cb)
From what I was able to debug, source of the problem is chunk size defined in the block meta json. For some reason it states smaller size than reality is, requested buffer is too small and store panics. I guess source of this will be compactor or receiver. We have also "historic stores" serving data created by Thanos v30 and older, there are no such panics, so I assume its combination of the store optimization together with something in receiver or compactor code not filing these values correctly in latest versions.
This will not fix source of this issue but at least fix store code which doesn't correctly check inputs.
Thanos, Prometheus and Golang version used:
Thanos: 0.32.5 Prometheus: 2.48
Object Storage Provider: AWS S3
What happened:
Using Thanos Store with Time Filter from time to time I'm it's crashing with the following message.
What you expected to happen:
Not crash
How to reproduce it (as minimally and precisely as possible):
Actually! I have no idea, this is only happening in one shard.
Full logs to relevant components:
Anything else we need to know: