How to find the setting for indexer.plasma.backIndex.evictSweepInterval

Couchbase Server
21

@varun.velamuri

Hi Varun, thanks a lot for your help so far. I was finally able to get the statistics when no mutation was happening on the indexes. I saw that total index data size is 84GB, but INDEX RAM usage is 120GB. So , could not figure out why the INDEX NODE is using so much of RAM. Also memory_total_storage shows 140200554496 bytes (140gb) for a 84GB index. Is it because of the back index size? But if that was the case, it should have been twice of 84GB, but it is not. I also think that there may be some memory calculation error, we now have a 256GB index node, but memory_total shows 270311915520. Should not this be equal to 256GB?

cpu_utilization: 20.333333333333332
index_not_found_errcount: 20.333333333333332
indexer_state: Active
memory_free: 104056897536
memory_quota: 246960619520
memory_rss: 132423802880
memory_total: 270311915520
memory_total_storage: 140200554496
memory_used: 140826920960
memory_used_queue: 0
memory_used_storage: 127307277434
needs_restart: False
num_connections: 8
num_cpu_core: 8
storage_mode: plasma
timestamp: 1632337274642875235
timings_stats_response: 66395 35210921943 138801384989183393
uptime: 14h3m39.928818265s
22

@rajib761 This is not expected. For all the indexed data, there will be an overhead of maintaining the indexed data but I think ~40G+ is a bit too much. Can you please share the complete output of api/v1/stats endpoint. You can redact the index names.

Also, can you please share the output of stats/storage endpoint of the indexer node after redacting index names. This helps us to validate and compute the overheads. We can take this forward from there.

Thanks,
Varun

23

Hi Varun, thanks for looking at it. We have changed our strategy now with indexes and are using partitioned index. Let me look at how this impacts the memory usage. I will get back to you based on the findings

24

@rajib761 From a memory usage or overhead perspective, it should be similar to a non-partitioned index as the number of documents indexed will not change. A partition index will distribute the index data over multiple nodes. So, overheads also get distributed.

25

Yes Varun, we have now shared the logs with Couchbase support team to investigate. memory usage still seems heavy