From: Matt Benjamin Date: Mon, 28 Mar 2022 16:50:11 +0000 (-0400) Subject: Update multisite-reshard.md X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=refs%2Fheads%2Fmattbenjamin-patch-1;p=ceph.git Update multisite-reshard.md --- diff --git a/src/doc/rgw/multisite-reshard.md b/src/doc/rgw/multisite-reshard.md index 32715290ebc1..1064052722d3 100644 --- a/src/doc/rgw/multisite-reshard.md +++ b/src/doc/rgw/multisite-reshard.md @@ -71,6 +71,7 @@ The distinction between *index layout* and *log layout* is important, because in * Full sync uses a single bucket-wide listing to fetch all objects. - Use a cls_lock to prevent different shards from duplicating this work. + - Casey pointed out that this substantially reduces parallelism of full sync, and Yuval has recently noticed that it may be easy to induce a scenario where a large amount of ingest into a new bucket is being handled by full sync (with reduced perf); could there be a way to prime new (and therefore empty) buckets for incremental sync on creation? * When incremental sync gets to the end of a log shard (i.e. listing the log returns truncated=false): - If the remote has a newer log generation, flag that shard as 'done' in the bucket sync status. - Once all shards in the current generation reach that 'done' state, incremental bucket sync can advance to the next generation.