On 7/23/19 4:30 PM, Waiman Long wrote: > On 7/22/19 8:46 AM, peter enderborg wrote: >> On 7/2/19 8:37 PM, Waiman Long wrote: >>> Currently, a value of '1" is written to /sys/kernel/slab/<slab>/shrink >>> file to shrink the slab by flushing all the per-cpu slabs and free >>> slabs in partial lists. This applies only to the root caches, though. >>> >>> Extends this capability by shrinking all the child memcg caches and >>> the root cache when a value of '2' is written to the shrink sysfs file. >>> >>> On a 4-socket 112-core 224-thread x86-64 system after a parallel kernel >>> build, the the amount of memory occupied by slabs before shrinking >>> slabs were: >>> >>> # grep task_struct /proc/slabinfo >>> task_struct 7114 7296 7744 4 8 : tunables 0 0 >>> 0 : slabdata 1824 1824 0 >>> # grep "^S[lRU]" /proc/meminfo >>> Slab: 1310444 kB >>> SReclaimable: 377604 kB >>> SUnreclaim: 932840 kB >>> >>> After shrinking slabs: >>> >>> # grep "^S[lRU]" /proc/meminfo >>> Slab: 695652 kB >>> SReclaimable: 322796 kB >>> SUnreclaim: 372856 kB >>> # grep task_struct /proc/slabinfo >>> task_struct 2262 2572 7744 4 8 : tunables 0 0 >>> 0 : slabdata 643 643 0 >> >> What is the time between this measurement points? Should not the shrinked memory show up as reclaimable? > > In this case, I echoed '2' to all the shrink sysfs files under > /sys/kernel/slab. The purpose of shrinking caches is to reclaim as much > unused memory slabs from all the caches, irrespective if they are > reclaimable or not. Well, SReclaimable counts pages allocated by kmem caches with SLAB_RECLAIM_ACCOUNT flags, which should match those that have a shrinker associated and can thus actually reclaim objects. That shrinking slabs affected SReclaimable just a bit while reducing SUnreclaim by more than 50% looks certainly odd. For example the task_struct cache is not a reclaimable one, yet shows massive reduction. Could be that the reclaimable objects were pinning non-reclaimable ones, so the shrinking had secondary effects in non-reclaimable caches. > We do not reclaim any used objects. That is why we > see the numbers were reduced in both cases. > > Cheers, > Longman >
