Whamcloud - gitweb
LU-6529 ldlm: reclaim granted locks defensively
To avoid ldlm lock exhausting server memory, two global parameters:
ldlm_watermark_low & ldlm_watermark_high are used for reclaiming
granted locks and rejecting incoming enqueue requests defensively.
ldlm_watermark_low: When the amount of granted locks reaching this
threshold, server start to revoke locks gradually.
ldlm_watermark_high: When the amount of granted locks reaching this
threshold, server will return -EINPROGRESS to any incoming enqueue
request until the lock count is shrunk below the threshold again.
ldlm_watermark_low & ldlm_watermark_high is set to 20% & 30% of the
total memory by default. It is tunable via proc entry, when it's set
to 0, the feature is disabled.
Signed-off-by: Niu Yawei <yawei.niu@intel.com>
Change-Id: I2fab39ac0ab6f269b7f1a40f3e08b8a51807cc69
Reviewed-on: http://review.whamcloud.com/14931
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Andreas Dilger <andreas.dilger@intel.com>
Reviewed-by: Bobi Jam <bobijam@hotmail.com>
Reviewed-by: Oleg Drokin <oleg.drokin@intel.com>
16 files changed: