Hi all
I have a quite big repository (approx 1Tb datastore, 100 Gb segmentstore) running on a AEM 6.1 instance. Configuration is:
. 1 author
. 3 publish
. 1 shared datastore (shared also between author and publish).
We are experimenting different issues on publish and author related to "tar optimization" (let's call it in 5.x style... at least it was working there).
Issues reported:
. 100% CPU on publish instances during normal daylight hours: checking logs and threaddumps it's quite clear that tar compaction is running
. Big increase in the segmentnodestore size (normally should be 100, we arrive quickly to 200 Gb and more... compaction not working at all)
Checking online documentation I have some problems understanding the situation:
. it seems that Adobe recommends (now) to perform ONLY offline compaction. Well... It's a bit hard to me to understand the suggestion... offline compaction on production servers? so I nee to think about shutting down servers in maintenance windows periodically? at least once a week I suppose... so this is *really* what Adobe suggests? I will have a lot to discuss with my sysadmin about this point... but above all... what was wrong with old tar optimization? so nice to complete and stop exactly when it was supposed to stop???
. ootb AEM 6.1 is defining a daily maintenance windows with online compaction tool running from 2 am to 5 am. But if Adobe recommends offline compaction, why ootb is setup? Checking online it's hard to understand how to remove the daily maintenance window. Is it enough to remove configuration node from crxde under: /libs/granite/operations/config/maintenance/granite:daily/granite:RevisionGC ? Is there something else?
. if a maintenance window is set to start at 2 am and to complete at 5 am.... can someone tell me the meaning of "END"? because it's clear that the compaction tool continues to work after 5 am and continues during the day with publish instances running at 100% CPU all day affecting performances.
Last thing not so clear (at least to me). Old, unreferenced, checkpoints need to be deleted before running offline compaction. Is there some resource online that is going to explain what such checkpoints are and how they are used? because in the error.log I also find AEM complaining about checkpoints but it's hard to find a single online note explaining what they are.
Very last: is it always TRUE that compaction is working ONLY at segment node level? I have a shared datastore so I know I need to perform datastore GC ONLY on authoring and not on publish. I would like to be sure that executing compaction (online or offline) is not triggering DS garbage collection as well. As far as I have understood this continue to work like in 5.5 with a mark & sweep process.
Thanks to everyone is going to help me better understand the new optimization logic... hoping to change my mind and thinking that compaction in 6.1 can be an improvement over previous tar optimization.
