We have a SQL Server 2012 R2 SP2 environment and Transactional Replication configured with 1 instance publishing all tables in a database via a single publication (circa 150 articles in the one publication). We have a separate distribution instance, and a third subscription instance. All articles in the publication are configured to copy all data, with no advanced filtering configured. It is also noted that Alerts were not configured in Replication Monitor.
We had an issue whereby, inexplicably, a table on the subscriber vanished (I use this term because I could not find any record in the log file of a DELETE statement being issued against that table - regrettably I wasn't in the position to query the distribution database for any such commands either).
In Replication monitor we saw that the article for the table in question was failing, but no others.
The issue was that no other articles were being replicated to the subscriber, and this was evident via stalled rowcounts on the subscriber database, but increasing rowcounts on the publishing database.
When we removed the offending article from the publication, all other articles started getting applied to the subscriber database - and we could observe both publication and subscription rowcounts increasing in lockstep.
My questions are:
1) in a single publication with many articles, are the articles applied to the subscription in a single batch - meaning that if a discreet transaction for one article fails, that all other articles will be rolled back with this transaction until it is cleared.
2) Why would Replication Monitor not start showing widespread warnings for every other table (article) that was not successfully being replicated to the subscriber?
3) Aside from configuring alerts in Replication Monitor (which has happened already) and a bespoke rowcount checking 'monitor', how could Transactional Replication be configured so that the failure of the publication of an obscure article doesn't fail all articles --> I note that creating multiple publications with smaller number of articles would limit the exposure, but from what I've observed, this would only limit the exposure to those articles in the same publication.
Many thnaks