Append / prepend: post-degradation recovery way


#1

Hello!
There is an interesting case (undersized, 3 nodes, 1CPU + 1G RAM):

  1. i used very long log via append() [increment + ~80 bytes per iteration]
  2. during tests, after some time, i faced with real performance degradation (as described here: http://developer.couchbase.com/documentation/server/4.1/developer-guide/raw-append-prepend.html#story-h2-2)
  3. log file size was 1.7Mb
  4. i destroyed and re-created all buckets i used.
    But:
  5. server still was under unusual heavy load (even with empty buckets)
  6. tests performance was lower then before
    I recovered with “full reinstallation” (and don’t have reasons now to investigate this case deeper; it’s easier to change data flow), but (supposing that this is known “feature” because of http://developer.couchbase.com/documentation/server/4.1/developer-guide/raw-append-prepend.html#story-h2-2 ), is there a “soft way” to recover from performance degradation ?
    And: are there other known cases, that could cause similar degradation ?

#2

Hi,
first please give more information:
1-machine spec: how many core, ram.
2-for cluster: how many ram.
3-for bucket: how many ram.
4-which version of couchbase?
but in totally as experience, [quote=“egrep, post:1, topic:8368”]

  1. server still was under unusual heavy load (even with empty buckets)
    [/quote] it occurred of OS ram used, it meant your OS could not did his job correct or complete.
    in fact, you may stop couchbase-server completely but RAM of OS may not free.
    after [quote=“egrep, post:1, topic:8368”]
  2. i destroyed
    [/quote] you may need check RAM usage to be sure who is using ram.

#3

@vhp1360,
this problem co-existed with https://issues.couchbase.com/browse/MB-19714,
so it’s hard to say now, what exactly had more impact on system behavior.


#4

thanks for your advise, but regarding ram issue, I guess i’m right. :slight_smile:
thanks.