Is this RAC or Exadata ? DRM node moves can be expensive , and if your
interconnect is dropping packets it becomes worse . Since you are getting cell
metrics, it implies Exadata . I would check gcs_server_processes and
high_priority_processes to make sure they are set properly .
Sent from my Atari 2600
On Nov 29, 2022, at 8:42 AM, Priit Piipuu <priit.piipuu@xxxxxxxxx> wrote:
Hi!
"gcs drm freeze in enter server mode" is specific to RAC. It seems to be
triggered by instance leaving or joining the cluster. Brownouts during
dynamic remastering is expected behaviour.
On Tue, 29 Nov 2022, 03:56 Lok P, <loknath.73@xxxxxxxxx> wrote:
Hello ,
It's version 19C of the oracle database. We got complaints from the
application team of sudden slowness and thus app traffic automatically
redirected to another active DB replica.While looking into the DB side, we
do observe high wait events from wait class 'Other' and 'Application' in
OEM. It lasted for exactly 2-3 minutes as the application team is
complaining about. And then fetching the "ash wait chain" and "ash top" ,
it's pointing to significantly high wait event like "reliable message"
(~75%), "enq: KO - fast object checkpoint","gcs drm freeze in enter server
mode" from same wait class 'other' and 'application'.
Below is the ASH top and ASH wait chain from the issue duration of ~5minutes
interval. And the select query which is coming on top. And table CB is
holding just ~41 rows in it.
https://gist.github.com/oraclelearner/44394ab8206fc7bd51041eb3d45bdf9f
And also the top query which is showing in "ash top" is a SELECT query and
it does have a FTS in it (which is suggesting why the checkpoint waits),
however, this select query doesn't have any plan change observed in it or
not any high execution. And also we are not seeing any specific blocking
session for this. The "wait chain" showing LMON/lock monitor process in one
of the lines also does point to some event like "Wait for msg sends to
complete ''. Not yet able to figure out how these are related to this issue.
Still trying to understand what could have caused such a scenario?
This SELECT query executes 100's of thousands of times in ~15minutes window.
So is it possible that this query(or say its underlying table CB) might have
thrashed out from buffer cache to disk and fetched data from disk rather
than cache for that time and this causing all these sort of issues? And
considering this table is very small (~1MB in size and holding just ~41
rows), should we consider putting it as inmemory by changing
"inmemory_clause_default" to 'BASE_LEVEL' and "INMEMORY_SIZE" to 16GB ?
