On Tue, 3 Jun 2025, Joshua Root wrote:
On 2/6/2025 21:41, Ryan Carsten Schmidt wrote:
On May 31, 2025, at 17:41, Sergey Fedorov wrote:
It is the second time in about 12 hrs that all ports are reindexing from
scratch. This should not be happening. Likely something messes up
timestamps on the server side.
The mprsyncup process is currently being run by Josh's server not mine. I
don't know how his process differs from mine. Something may need
fine-tuning on that server.
The timestamps in ports.tar look correct to me, and I haven't seen any
excessive indexing when syncing with rsync. The timestamps of course would
have changed when switching to a different server with a git checkout made at
a different time.
What I saw Friday night was that CI jobs were spending 6-12 minutes of
runtime and 40,000+ lines of logfile doing a portindex, before doing
anything specific to the port being tested. That sounds like a
manifestation of this problem.
Whenever git is involved, it's best to avoid depending on mtimes, since
git insists on clobbering them. That's unlikely to ever change, given
that Torvalds regards it as a feature.
Fred Wright
