Change Details

If two requests within 60 seconds of each other write to different sets of databases (and the second request's databases are not a superset of the first request's), ChronologyProtector will wind up waiting until timeout for databases that weren't written during the second request to be "updated" to the second request's timestamp. This causes slow (>5s) responses to subsequent requests until ChronologyProtector's cache expires (after 60 seconds of no further writes), both read and write. It looks like this is a long-standing bug that was made much more visible by the recent fix to actually use the cpPosTime cookie. See T182322#3823259 for a more detailed analysis.