X-Git-Url: https://gerrit.opnfv.org/gerrit/gitweb?a=blobdiff_plain;f=src%2Fceph%2Fdoc%2Fdev%2Fosd_internals%2Fpg_removal.rst;fp=src%2Fceph%2Fdoc%2Fdev%2Fosd_internals%2Fpg_removal.rst;h=d968eccc631d7be48a128895213383ab8047d2a9;hb=812ff6ca9fcd3e629e49d4328905f33eee8ca3f5;hp=0000000000000000000000000000000000000000;hpb=15280273faafb77777eab341909a3f495cf248d9;p=stor4nfv.git

diff --git a/src/ceph/doc/dev/osd_internals/pg_removal.rst b/src/ceph/doc/dev/osd_internals/pg_removal.rst
new file mode 100644
index 0000000..d968ecc
--- /dev/null
+++ b/src/ceph/doc/dev/osd_internals/pg_removal.rst
@@ -0,0 +1,56 @@
+==========
+PG Removal
+==========
+
+See OSD::_remove_pg, OSD::RemoveWQ
+
+There are two ways for a pg to be removed from an OSD:
+
+  1. MOSDPGRemove from the primary
+  2. OSD::advance_map finds that the pool has been removed
+
+In either case, our general strategy for removing the pg is to
+atomically set the metadata objects (pg->log_oid, pg->biginfo_oid) to
+backfill and asynronously remove the pg collections.  We do not do
+this inline because scanning the collections to remove the objects is
+an expensive operation.
+
+OSDService::deleting_pgs tracks all pgs in the process of being
+deleted.  Each DeletingState object in deleting_pgs lives while at
+least one reference to it remains.  Each item in RemoveWQ carries a
+reference to the DeletingState for the relevant pg such that
+deleting_pgs.lookup(pgid) will return a null ref only if there are no
+collections currently being deleted for that pg. 
+
+The DeletingState for a pg also carries information about the status
+of the current deletion and allows the deletion to be cancelled.
+The possible states are:
+
+  1. QUEUED: the PG is in the RemoveWQ
+  2. CLEARING_DIR: the PG's contents are being removed synchronously
+  3. DELETING_DIR: the PG's directories and metadata being queued for removal
+  4. DELETED_DIR: the final removal transaction has been queued
+  5. CANCELED: the deletion has been canceled
+
+In 1 and 2, the deletion can be canceled.  Each state transition
+method (and check_canceled) returns false if deletion has been
+canceled and true if the state transition was successful.  Similarly,
+try_stop_deletion() returns true if it succeeds in canceling the
+deletion.  Additionally, try_stop_deletion() in the event that it
+fails to stop the deletion will not return until the final removal
+transaction is queued.  This ensures that any operations queued after
+that point will be ordered after the pg deletion.
+
+OSD::_create_lock_pg must handle two cases:
+
+  1. Either there is no DeletingStateRef for the pg, or it failed to cancel
+  2. We succeeded in canceling the deletion.
+
+In case 1., we proceed as if there were no deletion occurring, except that
+we avoid writing to the PG until the deletion finishes.  In case 2., we
+proceed as in case 1., except that we first mark the PG as backfilling.
+
+Similarly, OSD::osr_registry ensures that the OpSequencers for those
+pgs can be reused for a new pg if created before the old one is fully
+removed, ensuring that operations on the new pg are sequenced properly
+with respect to operations on the old one.