Closed etiennedi closed 3 weeks ago
GOMEMLIMIT
GOGC
Everything is crash-resistant
After a while at least one of the pods is going to be in a crash loop logging something like the following:
panic: assignment to entry in nil map goroutine 600 [running]: github.com/weaviate/weaviate/adapters/repos/db/inverted.(*JsonPropertyLengthTracker).TrackProperty(0xc004b64240, {0xc003a6c290, 0x5}, 0x40a00000) /go/src/github.com/weaviate/weaviate/adapters/repos/db/inverted/new_prop_length_tracker.go:181 +0x3c8 github.com/weaviate/weaviate/adapters/repos/db.(*Shard).SetPropertyLengths(0xc003724b40, {0xc0063f1a40?, 0x4, 0x29cace0?}) /go/src/github.com/weaviate/weaviate/adapters/repos/db/shard_write_inverted_lsm.go:217 +0x8f github.com/weaviate/weaviate/adapters/repos/db.(*Shard).updateInvertedIndexLSM(0xc003724b40, 0xc0063f1880, {0xc0063a5310?, 0xe0?, 0x10?}, {0x0, 0x0, 0x0}) /go/src/github.com/weaviate/weaviate/adapters/repos/db/shard_write_put.go:299 +0x5ea github.com/weaviate/weaviate/adapters/repos/db.(*Shard).putObjectLSM(0xc003724b40, 0xc0063f1880, {0xc0063a5310, 0x10, 0x10}) /go/src/github.com/weaviate/weaviate/adapters/repos/db/shard_write_put.go:164 +0x725 github.com/weaviate/weaviate/adapters/repos/db.(*objectsBatcher).storeObjectOfBatchInLSM(0xc005405780, {0x1da6900, 0x29feb80}, 0x0?, 0xc0063f1880) /go/src/github.com/weaviate/weaviate/adapters/repos/db/shard_write_batch_objects.go:201 +0xd1 github.com/weaviate/weaviate/adapters/repos/db.(*objectsBatcher).storeSingleBatchInLSM.func1(0xf9, 0x0?) /go/src/github.com/weaviate/weaviate/adapters/repos/db/shard_write_batch_objects.go:173 +0x115 created by github.com/weaviate/weaviate/adapters/repos/db.(*objectsBatcher).storeSingleBatchInLSM in goroutine 283 /go/src/github.com/weaviate/weaviate/adapters/repos/db/shard_write_batch_objects.go:159 +0xc5
Discovered as part of #4125 investigations.
v1.23.7
Keep open until all #4125 changes are released, then re-evaluate. This should by fixed by those changes, too.
I think this is resolved
How to reproduce this bug?
GOMEMLIMIT
andGOGC
so OOMKills are likelyWhat is the expected behavior?
Everything is crash-resistant
What is the actual behavior?
After a while at least one of the pods is going to be in a crash loop logging something like the following:
Supporting information
Discovered as part of #4125 investigations.
Server Version
v1.23.7
Code of Conduct