hyc@symas.com wrote:

So even with this fix in HEAD, test058 still reports 2 errors for me. Are these patches really ready for test? (Or release?)

2 errors from test058 is currently to be expected, as are occational race errors. The number of race tests can be set with the RACE_TESTS environment variable (default 10). I have seldom seen it pass 100... Fixing this requires changing where syncrepl stores its contextCSN and how syncprov pics up its changes. I'll return on -devel with some thoughts I have about that, hopefully later today.

As of my recent additions, specifically ITS#5972 and ITS#5973, I have run with them in production since around 2.4.12, in a configuration similar to test058. 5972 I believe to be safe, but 5973 could need extra attention. Updating the csn as it does highly depends on all syncrepl received changes to be on the queue destined for the consumers before the csn update is queued, and being sent in the queued order. So, please take an extra look on that one!

Rein