Showing posts with label mode. Show all posts
Showing posts with label mode. Show all posts

Monday, March 12, 2012

Merge Replication limit running in continuous mode

Hi,
We have a customer site running merge replication with a
publisher/distributer and 12 subscribers. All the merge agents are running in
continuous mode. It seems that two of the agents are always failed, or don't
last long. Is this directly related to the "number of concurrent merge
processes" number? It looks like because there are 10 merge agents *always*
running, the other two don't get a look in. Is this correct?
We can increase this number to 20 and the site seems to run ok, but we've
been hearing rumours that any number past 10 concurrent merge processes gets
unstable. What's the MS position on this?
What's the recommended solution to this situation? Schedule the agents
instead of continuous mode?
What is the error message generated by these two always failing agents?
Do you have the "limit the number of concurrent merge processes to the
following" option checked and set to 10.
What happens is that for long running merge agents the agents will timeout
or deadlock themselves, so MS advised you to limit the number that run
concurrently.
If you do choose this option you have to configure your agents to wait a
longer time before timing out. To do this right click on your merge agents,
select agent properties, click on the Steps tab, and double click on Run
Agent. Place focus in the commands text box, press End, and press the space
bar. enter -StartQueueTimeout 120
This will have the effect of having these two failing merge agents of
waiting up to 2 minutes for the first 10 merge agents in the queue to finish
their work.
I am not sure what the rational for 10 was. The fact it seems to work fine
when you bump it up to 20 seems to indicate that this throttling is not
optimal for you. I suspect that you did not use StartQueueTimeout option.
Running the agents in continuous mode will have no effect. The agents poll
by default every minute, so after the agent completes it will go to sleep
for a minute, and any agents in the queue will have their timeslice.
Hilary Cotter
Looking for a SQL Server replication book?
http://www.nwsu.com/0974973602.html
Looking for a FAQ on Indexing Services/SQL FTS
http://www.indexserverfaq.com
"Sydney" <hsc@.newsgroup.nospam> wrote in message
news:5C49D704-C649-41FA-8686-3EA92151867F@.microsoft.com...
> Hi,
> We have a customer site running merge replication with a
> publisher/distributer and 12 subscribers. All the merge agents are running
> in
> continuous mode. It seems that two of the agents are always failed, or
> don't
> last long. Is this directly related to the "number of concurrent merge
> processes" number? It looks like because there are 10 merge agents
> *always*
> running, the other two don't get a look in. Is this correct?
> We can increase this number to 20 and the site seems to run ok, but we've
> been hearing rumours that any number past 10 concurrent merge processes
> gets
> unstable. What's the MS position on this?
> What's the recommended solution to this situation? Schedule the agents
> instead of continuous mode?
>

Friday, March 9, 2012

Merge replication ExchangeType question.

I have an application that will have very limited functionality in a
disconnected mode. I plan on recording operations in the one area of our
application where we allow inserts in disconnected mode and updating the
publisher through custom code that basically replays what the user did in
disconnected mode. Then using merge replication in dowload only mode to pull
any updates back to the subscriber.
My question is do I need to delete the rows that were inserted in
disconnected mode on the subscriber before starting the merge process. I am
concerned that the publisher will see the inserts from the custom code and
attempt an insert on the subscriber even though the row already exists there.
Is merge replication smart enough (in download only mode) to delete the rows
that were added in disconnected mode on the subscriber (since they don't
exist on the publisher) before inserting the rows from the publisher? To put
it another way, does merge replication try to completely synchronize data on
both the subscriber and the publisher by forcing the subscriber to have the
same data as the publisher in download only mode?
I think my last post may have been confusing.
Can the data be modified or inserted on the subscriber but then the
subscriber be forced to synchronize completely with the publisher's copy of
the database. In other words, any data created on the subscriber is purged
during the merge process instead inserted into the publisher?
|||I'd have to test, but I think you'll have to remove these rows yourself. You
can identify them by matching the guids in msmerge_contents, and this could
be an extra step in the merge agent's job.
Rgds,
Paul Ibison SQL Server MVP, www.replicationanswers.com
(recommended sql server 2000 replication book:
http://www.nwsu.com/0974973602p.html)
|||Yes, you will need to delete the rows on the subscriber, otherwise you will
get a conflict when the row moves from the publisher to the subscriber.
Hilary Cotter
Looking for a SQL Server replication book?
http://www.nwsu.com/0974973602.html
"mspradley" <mspradley@.discussions.microsoft.com> wrote in message
news:8F9DE6A4-B8F1-45D3-843C-4EA6F5F6BF94@.microsoft.com...
> I have an application that will have very limited functionality in a
> disconnected mode. I plan on recording operations in the one area of our
> application where we allow inserts in disconnected mode and updating the
> publisher through custom code that basically replays what the user did in
> disconnected mode. Then using merge replication in dowload only mode to
pull
> any updates back to the subscriber.
> My question is do I need to delete the rows that were inserted in
> disconnected mode on the subscriber before starting the merge process. I
am
> concerned that the publisher will see the inserts from the custom code and
> attempt an insert on the subscriber even though the row already exists
there.
> Is merge replication smart enough (in download only mode) to delete the
rows
> that were added in disconnected mode on the subscriber (since they don't
> exist on the publisher) before inserting the rows from the publisher? To
put
> it another way, does merge replication try to completely synchronize data
on
> both the subscriber and the publisher by forcing the subscriber to have
the
> same data as the publisher in download only mode?