HumanCellAtlas / logs

Infrastructure for centralizing and indexing logs

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Kibana in `dev` returns duplicate hits for a single log record

hannes-ucsc opened this issue · comments

One log record:

screen shot 2018-09-26 at 8 45 11 pm

Nine hits:

screen shot 2018-09-26 at 8 48 22 pm

@hannes-ucsc are you still seeing this behavior? I can't seem to replicate this.

Yes, occasionally. How did you attempt to reproduce it?

@hannes-ucsc I dug through some log lines similar to the ones you showed and tried to find duplicates

@mweiden I saw this at considerable scale yesterday. I was expecting 130k log messages for a particular Azul operation but got 160k. Here is a query showing such a message that was indexed in quadruplicate:

screen shot 2018-12-04 at 1 25 23 pm

The key takeaway is that we have four distinct values for _id (the ES document identifier) but the same value for @id (the CloudWatch log record identifier IIRC). Can we use @id to populate _id?

I think this may have something to do with an OOM issue we're seeing in the dev deployment...

OOM report

REPORT RequestId: 04f76c37-fca4-11e8-a7ec-e52492b51ce1	Duration: 5983.41 ms	Billed Duration: 6000 ms Memory Size: 1024 MB	Max Memory Used: 1024 MB	

This is retried

RequestId: 04f76c37-fca4-11e8-a7ec-e52492b51ce1 Process exited before completing request
...
RequestId: 04f76c37-fca4-11e8-a7ec-e52492b51ce1 Process exited before completing request
...
RequestId: 04f76c37-fca4-11e8-a7ec-e52492b51ce1 Process exited before completing request