Skip to content

Commit 540b055

Browse files
authored
Merge pull request ceph#59168 from zdover23/wip-doc-2024-08-12-cephfs-cache-configuration
doc/cephfs: improve cache-configuration.rst Reviewed-by: Anthony D'Atri <[email protected]> Reviewed-by: Patrick Donnelly <[email protected]>
2 parents cd51a12 + aa3bdae commit 540b055

File tree

1 file changed

+6
-7
lines changed

1 file changed

+6
-7
lines changed

doc/cephfs/cache-configuration.rst

Lines changed: 6 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -237,12 +237,11 @@ all cached caps (the actual recall value is ``num_caps -
237237
mds_min_caps_per_client(100)``).
238238

239239
Under certain circumstances, many "recall caps" requests can be sent so quickly
240-
that the "mon warning limit" exceeded, and the "clients failing to respond to
241-
cache pressure" message can be triggered. If the client does not release the
242-
caps fast enough, the MDS repeats the "recall caps" request one second later.
243-
This means that the MDS will send "recall caps" again and again. The "total"
244-
counter of "recall caps" for the session will grow and grow, and will
245-
eventually exceed the "mon warning limit".
240+
that the health warning is generated: "clients failing to respond to cache
241+
pressure". If the client does not release the caps fast enough, the MDS repeats
242+
the "recall caps" request one second later. This means that the MDS will send
243+
"recall caps" again and again. The "total" counter of "recall caps" for the
244+
session will grow and grow, and will eventually exceed the "mon warning limit".
246245

247246
A throttling mechanism, controlled by the ``mds_recall_max_decay_threshold``
248247
parameter (126K by default), is available for reducing the rate of "recall
@@ -258,7 +257,7 @@ Example Scenario
258257
Here is an example. A client is having 20k caps cached. At some moment the
259258
server decides the client is inactive (because the session's ``cache_liveness``
260259
value is low). It starts to ask the client to release caps down to
261-
``mds_min_caps_per_client`` value (100 by default). For this every seconds it
260+
``mds_min_caps_per_client`` value (100 by default). Every second, it
262261
sends recall_caps asking to release ``caps_num - mds_min_caps_per_client`` caps
263262
(but not more than ``mds_recall_max_caps``, which is 30k by default). A client
264263
is starting to release, but is releasing with a rate of (for example) only 100

0 commit comments

Comments
 (0)