Dgraph repeatedly gets stuck on the same DQL transaction and aborts it after a while.
The transaction itself searches for nodes and adds the results to an UID list. The number of found UIDs is ~30-70k.
On the Go client I get: Transaction has been aborted. Please retry. I can trigger this behavior repeatedly for the same transaction, which never succeeds.
Here the alpha logs when the transaction gets aborted:
I0127 10:02:02.045457 1 draft.go:1448] Found 1 old transactions. Acting to abort them.
I0127 10:02:02.046778 1 draft.go:1409] TryAbort 1 txns with start ts. Error: <nil>
I0127 10:02:02.046826 1 draft.go:1432] TryAbort selectively proposing only aborted txns: txns:<start_ts:17134 >
I0127 10:02:42.074992 1 draft.go:1451] Done abortOldTransactions for 1 txns. Error: <nil>
I0127 10:05:57.335150 1 log.go:34] Block cache metrics: hit: 18573167 miss: 8864772 keys-added: 1043679 keys-updated: 1620634 keys-evicted: 944695 cost-added: 10399146412 cost-evicted: 9701232748 sets-dropped: 2516511 sets-rejected: 3683333 gets-dropped: 12511616 gets-kept: 14463488 gets-total: 27437939 hit-ratio: 0.68
I0127 10:08:02.043664 1 draft.go:1448] Found 1 old transactions. Acting to abort them.
I0127 10:08:02.045260 1 draft.go:1409] TryAbort 1 txns with start ts. Error: <nil>
I0127 10:08:02.045307 1 draft.go:1432] TryAbort selectively proposing only aborted txns: txns:<start_ts:17178 >
I0127 10:08:55.890128 1 draft.go:1451] Done abortOldTransactions for 1 txns. Error: <nil>
The following screenshot shows the metrics while the transaction is being processed and aborted multiple times.
This seems to a problem multiple people have:
Dgraph Metadata
dgraph version
v20.11
I running 1 zero and 1 alpha on the same machine (32 cpus, >150 GB RAM). The db is 20-30 GB big after a fresh import.