On 03/29/2018 02:15 PM, Gregory Farnum wrote:
On Thu, Mar 29, 2018 at 2:08 PM, Josh Durgin <jdurgin@xxxxxxxxxx> wrote:
Another aspect that hasn't come up yet is keeping a strict limit on the
log size, so that we can keep a bounded ring buffer instead of growing
it extensively and incurring extra overhead during recovery/backfill.
Right now we set min_last_complete_ondisk based on the
acting_recovery_backfill set, so we end up not trimming the log during
backfill and async recovery.
Is there any reason not to trim the logs on the acting set at least? The
async recovery and backfill shards need the longer log to stay
contiguous with the acting set/not restart backfill on interval change,
but among the acting set I don't see what issues this would cause.
If we do that, any node which goes down temporarily has a good chance
of no longer being contiguous, and having to shift from recovery to
backfill, or restart backfill.
Ah right, that does affect any node still. We could cap it for all nodes
at a higher limit though, so we don't have unbounded memory growth.
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
