Open sivesh1989 opened 1 year ago
We are experiencing the same error. I think it is because memcached has to assemble the response in memory from the data on disk but the memcached has not enough memory. If you do stats on the memcached there is a stat about extstore oom. We probably need to give more memory to memcached. I saw this in memcached discussion in discord, I can link it here when I find it back
We are experiencing the same error. I think it is because memcached has to assemble the response in memory from the data on disk but the memcached has not enough memory. If you do stats on the memcached there is a stat about extstore oom. We probably need to give more memory to memcached. I saw this in memcached discussion in discord, I can link it here when I find it back
hi. Have you solved this problem?
yes, it is working fine. Make sure to read the wiki https://github.com/memcached/memcached/wiki/ConfiguringLokiExtstore, the maintainer of memcached did analysis and came up with recommendations. Especially the setting related to ext store made a big difference. Another take-away is that memcached with loki scales better horizontally (better a lot of small memcached than a single big)
yes, it is working fine. Make sure to read the wiki https://github.com/memcached/memcached/wiki/ConfiguringLokiExtstore, the maintainer of memcached did analysis and came up with recommendations. Especially the setting related to ext store made a big difference. Another take-away is that memcached with loki scales better horizontally (better a lot of small memcached than a single big)
Nice, Thank you for your help. I'm a bit confused. memcache Extstore refers to memcachedFront or memcachedChunks ? My loki cluster is deployed by help loki distributed, and I set the memcache start option -m 6000 -I 2m -o ext_path=/disk/extstore:500G,ext_wbuf_size=32,ext_threads=10,ext_max_sleep=10000,slab_automove_freeratio=0.10,ext_recache_rate=0
, I encounter an error: 'Illegal suboption "(null)"'. Our daily log generation is around 6T+ and stored in S3, and the query efficiency is very low. It's quite problematic.
the extstore is especially usefull for the chunk cache, because it is big if you want to keep multiple days, so disk can be more cost effective then RAM. But you need decent disk performance, the ext_threads is all dependent on the disk speed I think. The query result cache is small and does not need an extstore. You should head on to the slack grafana/loki channel with your question. Once I am back in the office I can give you the command we use for memcached ( I surely don't have the error illegal suboption !)
the extstore is especially usefull for the chunk cache, because it is big if you want to keep multiple days, so disk can be more cost effective then RAM. But you need decent disk performance, the ext_threads is all dependent on the disk speed I think. The query result cache is small and does not need an extstore. You should head on to the slack grafana/loki channel with your question. Once I am back in the office I can give you the command we use for memcached ( I surely don't have the error illegal suboption !)
Okay, I have already posted my question on the grafana/loki channel and look forward to your response. Thank you
@waney316 Are you solved this problem ?
I`m facing this same issue.
Hi,
While performing read path greater than 2 hrs data we observe below error,
2023-09-25T06:46:45.523875277Z stderr F ts=2023-09-25T06:46:45.442137322Z caller=spanlogger.go:86 user=xxx level=warn msg="error fetching from cache" err="memcache: unexpected line in get response: \"SERVER_ERROR out of memory writing get response\r\n\""
Querier memory utilization is normal around 10%( Limits: 50GB) Memcache replica count 40 Querier Replica count 50
Regards, Sivesh Kumar R