Hello
- SQL Server 2008 R2
- 400Gb Database running in SQL 2000 compatability mode
I am pulling my hair out somewhat with replication. We have been using it for years without making too many changes to it. Four publications with between three and four subscribers to each publication. Latency has been around 2 seconds to 5 minutes as the years have passed. Every six months or so we probably recreate the publications and subscriptions.
This last week the log reader has been having issues. The latency issues are between the publisher and the distributor where latency is up currently at 1h46m, between the distributor and the subscriber latency is a few seconds.
Where do I go to troubleshoot this? So much of it is "under the hood". I do not know what stats to look at, let alone where or how to collect them.
We did make some changes that resulted in one of the publications breaking. We added a varchar(max) field to our database but did not set the option to "Convert MAX datatypes to NTEXT" which seems to be the requirement. (With this version of the database varchar(max) is stored as text and this option has to be set to replicate such fields around.)
We couldn't get replication to get going again so had to shut down the website (has about 9,000 visitors at any one time), to recreate the publications and subscriptions. This took about an hour and eventually the latency returned to normal.
Upon checking this morning though it was back up at 1 hour. After reading we removed an article that was replicating a lot of records with text fields and again latency seemed to return to normal. We looked at other ways of replicating this information but in the meantime the latency crept back up again to where we are now. Using tracer tokens also confirms the latency and where it is between.
I have spent days trying to figure this out whilst balancing the need for the Site to be available to the membership. Trying to do any publication creation or snapshot building pretty much fails if we allow the users to be on the Site. Timeouts, deadlocks ... that kind of thing.
We are performing transaction log backups every 30 minutes. This causes truncation delays as the reader is struggling to read all of the commands/transactions it needs.
CPU usage on the distributor is very low, single %s. Plenty of disk space.
How can I find out which updates/commands are generating so much activity that the log reader cannot keep up?
Many thanks in advance.
Alan