Stefan,

I have been running logging and I think I have found the problem however I do not know how to resolve it.

Here is where it is going wrong:

19-May-2009 04:02:48 [] {NSS} MIRROR: ID21268: Local entry submitted

19-May-2009 04:07:34 [] {AMC} MIRROR: send entry #31056

It seems that when it has finished replication on one logbook there is a significant time before the next logbook replication starts. During this time the server is not avalable. I have noticed that the time between ending one logbook and starting the next differs betwee 2 and 5 minutes. Again the server is not available when this happens.

Do you have any idea?

Cheers,

Lance

Hi,

I need more information to narrow down the problem:

- does this only happen on automatic mirroring or also when you do a manual synchronize?

- does this happen on both sides, like when you make the "other" elog server the master?

- if the server is inresponsive, does the CPU on that machine go to 100% or to 0%?

- do you have very long attachments in your logbooks?

- do you have the same problem on a "tiny" logbook like it comes with the distribution?

- are you using any Apache proxy in between?

I'm afraid that in the end you have to debug this yourself, since it will be very hard for me to reproduce exactly your problem (unless you send me all your files). 

Cheers,

  Stefan

Stefan,

Thanks for the reply.

This happens on automatic mirroring and by manual sync. However only the site initializing the mirror is locked out the remote seems to still be able to function.

The CPU jumps from very little usage to 50%+ being used by elogd.exe as soon as you start the mirroring/sync process

I have attached a file that that is in three parts and its pretty big. When I start up the elogd -v it takes over two minutes to scroll through hundreds of  files. I have attached the last of those entries in the first part of the attached PDF, the second part of the PDF shows a manual sync and the third part shows the same sync on the same logbook a few mins later. It seems to take about 3 minutes even when there has been no new log entries. In addition if you are mirroring more that one log book through the automated cron job it can take about 3-5 mins before the second logbook starts its replication. I have also added a screenshot of the completed replications on both runs.

If there is a way to redirect the output of the cmd window when running elogd -v I would capture all the data for you but the standard redirect ">> elog.txt" only creates a blank file.

We are running several logbooks and it does look like the smaller logbooks still take several minutes to start up. I have attached the PMCLogfile and if you look between the NSS and the AMC replications on any day there seems to be a 3 min gap between one book finishing and another starting.

We are not using Apache prox in between.

I am not a programmer but I can follow instruction, if you need anything else let me know.

Stefan this has been driving me nuts for a while now so any help you can give would be more than appreciated.

Cheers,

Lance

Sorry my late reply but I'm pretty busy these days...

I don't have a clear solution, just a few thoughts:

- Network handling has benn improved recently, so I propose you first upgrade to Version 2.7.7

- Looking at your sync logs, I see many lines of the form

19-Jun-2009 15:41:05 [lance@127.0.0.1] {NSS} MIRROR change entry #1095 to #23357
19-Jun-2009 15:41:05 [lance@127.0.0.1] {NSS} DELETE entry #1095
19-Jun-2009 15:41:05 [lance@127.0.0.1] {NSS} MIRROR send entry #23357

this indicates that you add entries to both logbooks (with ID 1095) in this case. Then elog has a problem, since you have new entries with ID 1095 on both sides. So the only solution is to re-submit the entry #1095 on the source logbook as a new entry (with ID #23357 in this case), delete the old one and then submit the new one. This happens very often, which takes quite some time. Mirroring mainly makes sense if there is one active logbook where new entries gets submitted, and the second logbook is mainly as backup and read-only. Then mirroring is very effective. If you submit on both sides very heavily new entries, the merge process is quite complicated.

- If nothing has changed on both sides and you still have heavy synchronization work, it means that both logbooks kind of became inconsistent, and elog tries to sort that out. So a good starting point is to manually copy all xxxxxxa.log files from one side to the other, thus ensuring both logbooks are 100% identical. Then restart both elogd servers, issue a manual synchronization, and make sure it reports back to you that everything is identical.

Hope some of this helps,

  Stefan
 

   $Id: elogd.c 1660 2006-02-17 19:48:12Z ritt $