Milvus Slow Insert

[wilsonlimaneto]

Hey, I am setting up a Milvus (2shards+Minio) instance within AWS Fargate (which allowed me to test several pods of several sizes) and Postres RDS but it seems I am having in any ways the same slow insertion speed when I use write_documents: avg of 30k 1024 vectors took 3 hrs.

I am running out of clues, so I came here for help, if anybody had faced that or could give me some more ideas.

Thanks

[masci]

Not sure how much you can share about your deployment but this is pretty hard to answer without knowing it in deep detail.

I would start ruling out possible culprits:

• benchmark Postgres insertions in the same table created by Haystack from the same container
• if postgres is fine, investigate where time is spent, possibly timing Milvus operations alone
• ensure the cluster has enough resources or at least there are no spikes in the basic health metrics

I think the best course of action would be using some APM technology to instrument your Haystack code (I'm a big Datadog fan and I can definitely recommend it), that would give you an answer to all the points above at once.