How to optimize InfluxDB 1.X configuation using TSI1 index-version

Store
1

Hello,

I’m working with a DB with index-version = inmem configuration. I want to migrate it to TSI, but I’m not sure how to configure or tunning the parameters related with the new index version.

I have been searching on OOM related issues and got the following configuration, but I’m still no able to get when reduce or increase each of those values, so there is a list with some considerations, according to the issues comments:

index-version = “tsi1”
max-concurrent-compactions = 0 # set to 1 to control memory compactions
cache-max-memory-size = “1g" # lower to reduce cache
cache-snapshot-memory-size = "25m"
compact-throughput = "48m"  # lower to reduce memory consumption
compact-throughput = "48m" # lower to reduce memory consumption
max-index-log-file-size = 131072 # lower to ensure index is flushed
series-id-set-cache-size=0 # set by new PR https://github.com/influxdata/influxdb/pull/18286

Assuming the DB schema is OK, number of series is almost fixed and there is resources enough:
I would like to ask if:

  • The above considerations are true?
  • The default values provided by InfluxDB are the optimal ones or we should change when we see some memory/resources problems?
  • There are some guidelines or troubleshooting to apply a configuration change depending on the case?

Thanks,
Regards!

2

Hello @sbengo,
I asked a Sales Engineer to double check, but those settings are optimized, so I would try running TSI as is with those default settings. The only thing you could maybe change is cache-snapshot-memory-size. If you a lot of RAM, you could increase that to 50–but only if you notice compaction spikes. I would ask that you just try the default settings, and if you experience any problems come back and share them.
Thanks!