While looking at bug 63616, I tried limiting $wgFlowDefaultLimit and $wgFlowMaxLimit on my local machine to 1.
Topics displayed should be exactly the same, they should only load in different batches (per 10 or per 1).
Yet, I found that when loading per 1, there was a topic I could not find when loaded per 1. Conversely, when loaded per 1, there was a different topic that was not shown when loaded per 10.
Version: unspecified
Severity: normal
On a second try (again with flushed cache), the entry missing from the list loaded per 10, was there. And there was another one missing...
It seems like we're losing topics here & there.
I've done some minor testing without cache ($wgFlowUseMemcache = false) and couldn't find any dropped posts there.
Since I coulnd't find any consistency in the posts dropped earlier, it could just be that my local # messages sample size was too small, or just dumb luck that all of them appeared to exist now; however, it's an indication that this issue may not exist without cache, and that it's related to caching (and not related to pagination, filtering out suppressed comments, ...)
https://gerrit.wikimedia.org/r/#/c/124342/ seems to also have solved this, after some quick initial testing. Still need to figure out why that one was causing bad cache data, though.
Change 124645 had a related patch set uploaded by Matthias Mullie:
Fix DB-to-cache pipeline