From 19fcec84d8d7d21e796c7624e521b60d28ee21ed Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sun, 7 Apr 2024 20:45:59 +0200 Subject: Adding upstream version 16.2.11+ds. Signed-off-by: Daniel Baumann --- doc/dev/osd_internals/pgpool.rst | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) create mode 100644 doc/dev/osd_internals/pgpool.rst (limited to 'doc/dev/osd_internals/pgpool.rst') diff --git a/doc/dev/osd_internals/pgpool.rst b/doc/dev/osd_internals/pgpool.rst new file mode 100644 index 000000000..45a252bd4 --- /dev/null +++ b/doc/dev/osd_internals/pgpool.rst @@ -0,0 +1,22 @@ +================== +PGPool +================== + +PGPool is a structure used to manage and update the status of removed +snapshots. It does this by maintaining two fields, cached_removed_snaps - the +current removed snap set and newly_removed_snaps - newly removed snaps in the +last epoch. In OSD::load_pgs the osd map is recovered from the pg's file store +and passed down to OSD::_get_pool where a PGPool object is initialised with the +map. + +With each new map we receive we call PGPool::update with the new map. In that +function we build a list of newly removed snaps +(pg_pool_t::build_removed_snaps) and merge that with our cached_removed_snaps. +This function included checks to make sure we only do this update when things +have changed or there has been a map gap. + +When we activate the pg we initialise the snap trim queue from +cached_removed_snaps and subtract the purged_snaps we have already purged +leaving us with the list of snaps that need to be trimmed. Trimming is later +performed asynchronously by the snap_trim_wq. + -- cgit v1.2.3