We have a mature replication system to move data from OLTP to BI. It's been working for years with relatively few troubles. This week we've run into the same problem two days in a row with snapshot publications.
The first error was BCP failed to insert due to a NULL constraint issue on the target. Interestingly, the source data has a non null constraint and therefore none of the source data has nulls. Note, that this publication hasn't changed and neither has the source or destination schemas. It worked one day, failed the next. And to fix it, we simply reinitialized and reapplied the snap.
The next day, a similar issue - different publication, slightly different error (duplicate primary key (0,NULL)). It's worked every day for months before and to fix it, we reinit the snap and it works.
Looking deeper we see the bcp file that it fails on is MUCH larger than the other files in the group (1Mb vs 1K). When we open this file, it has about 1K of useful data followed by 999K of NULL characters. Anyone have a clue as to why this might happen? Why does it work one day, then fail, then work the next time we try it?