Importer record file parsing duration spikes when ingesting extremely large record files with > 50k HCS topic messages in perfnet, temporarily disabling autovacuum on topic_message table can reduce the time to less than 2 seconds, however with certain cloudsql instances it doesn't help at all.
On the other side, the current autovacuum settings for most insert only tables are tuned for balanced writes and reads, we need to investigate the impact to database reads if we need to change the settings to favor writes over reads.
Steps to reproduce
Restore the database from a recent backup before the TPS spike
Run importer and observe parsing duration when it starts to ingest large record files
We now believe this is due to a CPU config change that did not grant the importer enough CPU to handle the transaction spike in testnet. This has since been re-configured correctly.
Description
Importer record file parsing duration spikes when ingesting extremely large record files with > 50k HCS topic messages in perfnet, temporarily disabling autovacuum on
topic_message
table can reduce the time to less than 2 seconds, however with certain cloudsql instances it doesn't help at all.On the other side, the current autovacuum settings for most insert only tables are tuned for balanced writes and reads, we need to investigate the impact to database reads if we need to change the settings to favor writes over reads.
Steps to reproduce
Additional context
No response
Hedera network
other
Version
v0.95.0-SNAPSHOT
Operating system
None