Too many replication jobs?

  • I have a server acting as the publisher with 18 publications.

    Each publication has 8 subscribers. That's 144 jobs with just the distribution agent alone.

    I added a new publication and only two of the subscriber jobs will run, and the snapshot will not complete.

    We have had issues running many processes under the same service account, so our solution would be to add additional service accounts.

    We did that, and this time it did not work. Any ideas? Is there a hard limit in Windows 2003 that I might be running into as far as job stream?

  • the distribution agent runs as an OS process ... check the running processes in task manager ..... and I believe there is an upper limit to amount of such processes at the Os level, which has fix where you have to increase the limit in registry .... I read an article about it a while ago .... but could not find it right now 🙁

    -------------------------------------------------
    -Amit
    Give a man a fish and he'll ask for a lemon. Teach a man to fish and he wont get paged on weekends !! :w00t: - desparately trying to fish [/size]

  • Yes - I remember making some registry setting changes via a call from Microsoft Premier Support. After making those changes I was able to run more replication processes, but all sorts of other issues happened so we ended up having to revert those settings.

    I do believe that I'm running into a hard limit of OS processes.

  • You could change the subscriptions so that they used shared agents instead of independent agents (which is the default). Independent agents give you better performance than shared agents but clearly, there are other issues.

  • I can't find any support documentation to switch to shared agents. I'm assuming you would have to do that from T-SQL?

  • You have to drop the subscriptions, then go to publication properties, subscription options, top option is independent agent. Set that to false. Sadly you can't change it unless you have no subscriptions. 🙁

    David

    @SQLTentmaker

    “He is no fool who gives what he cannot keep to gain that which he cannot lose” - Jim Elliot

  • Thank you so much. I could not find that anywhere on the net or in Books online.

  • SQL DBA 808 (11/11/2009)


    Thank you so much. I could not find that anywhere on the net or in Books online.

    I only know it as we have had to work through the same problem that you have. We try to structure our publications so that those that can incur some latency are in one publication or if they are multiple publications they have shared agents. Those that require as little latency as possible or where there is a lot of replication traffic we will use an independent agent. We have been able to keep things pretty low after a lot of rework.

    One note, if you can quiesce the inserts, updates, deletes to the published table you can drop your subscriptions, change to shared agent and then add the subscription back and specify "do not initialize" during the subscription creation process. Saves the pain of the snapshot if this is a big table.

    Hope this helps.

    David

    @SQLTentmaker

    “He is no fool who gives what he cannot keep to gain that which he cannot lose” - Jim Elliot

  • I made the switch, but it still didn't give enough back to the system.

    I will have to change more to shared, or expand the desktop heap settings it appears.

  • So you have one distributor but how many distribution databases do you have?

    I ran into a problem where the distributor didn't look like it was stressed but the outstanding transactions were growing and growing.

    Switching one or two subscriptions to PULL subscriptions helped a bit but the thing that resolved it was having separate distribution databases per publisher.

  • we have 18 publications, with one distribution database.

    I can not pinpoint the issue because this isn't happening if I set the agents to run as the sql server agent account. When I use a domain account is when it fails replicate. I would think it was a permission issue, but that same domain user is running several other publications.

    I was running into an issue where the sql agent account could not run any more processes, so I switched new publications to a domain account (which solved that). But now it's the other way around, the sql agent user works and the domain user doesn't.

    After struggling two days with this, I am seriously doubting my replication skills.

  • Additionally, after adding more than one subscriber to the new publication the event viewer blows up with Distrib.exe can not initialize application.

    I think we are hitting the ceiling of our Desktop Heap. I read some documents on Microsoft related to it.

    We made a change to the registry setting, but that REALLY blew up replication and all sorts of strange things started stalling. We reverted back as this my production system. and replicating data is basically what pays my rent 🙂

Viewing 12 posts - 1 through 11 (of 11 total)

You must be logged in to reply to this topic. Login to reply