Message ID | 20210914072938.6440-1-songmuchun@bytedance.com (mailing list archive) |
---|---|
Headers | show |
Series | Optimize list lru memory consumption | expand |
On Tue, Sep 14, 2021 at 03:28:22PM +0800, Muchun Song wrote: > So we have to convert to new API for all filesystems, which is done in > one patch. Some filesystems are easy to convert (just replace > kmem_cache_alloc() to alloc_inode_sb()), while other filesystems need to > do more work. From what I can tell, three are 54 file systems for which it was a trivial one-line change, and two (f2fs and nfs42) that were a tad bit more complex. > In order to make it easy for maintainers of different > filesystems to review their own maintained part, I split the patch into > patches which are per-filesystem in this version. I am not sure if this > is a good idea, because there is going to be more commits. What I'd actually suggest is that you combine all of the trivial file system changes into a single commit, and keep the two more complex changes for f2fs and nfs42 in separate commits. Acked-by: Theodore Ts'o <tytso@mit.edu> ... for the ext4 related change. - Ted
On Wed, Sep 15, 2021 at 4:23 AM Theodore Ts'o <tytso@mit.edu> wrote: > > On Tue, Sep 14, 2021 at 03:28:22PM +0800, Muchun Song wrote: > > So we have to convert to new API for all filesystems, which is done in > > one patch. Some filesystems are easy to convert (just replace > > kmem_cache_alloc() to alloc_inode_sb()), while other filesystems need to > > do more work. > > From what I can tell, three are 54 file systems for which it was a > trivial one-line change, and two (f2fs and nfs42) that were a tad bit > more complex. Definitely right. Thanks for your clarification. > > > In order to make it easy for maintainers of different > > filesystems to review their own maintained part, I split the patch into > > patches which are per-filesystem in this version. I am not sure if this > > is a good idea, because there is going to be more commits. > > What I'd actually suggest is that you combine all of the trivial file > system changes into a single commit, and keep the two more complex > changes for f2fs and nfs42 in separate commits. Got it. Will do in the next version. > > Acked-by: Theodore Ts'o <tytso@mit.edu> Thanks. > > ... for the ext4 related change. > > - Ted >
On Tue, Sep 14, 2021 at 03:28:22PM +0800, Muchun Song wrote: > We introduced alloc_inode_sb() in previous version 2, which sets up the > inode reclaim context properly, to allocate filesystems specific inode. > So we have to convert to new API for all filesystems, which is done in > one patch. Some filesystems are easy to convert (just replace > kmem_cache_alloc() to alloc_inode_sb()), while other filesystems need to > do more work. In order to make it easy for maintainers of different > filesystems to review their own maintained part, I split the patch into > patches which are per-filesystem in this version. I am not sure if this > is a good idea, because there is going to be more commits. > > In our server, we found a suspected memory leak problem. The kmalloc-32 > consumes more than 6GB of memory. Other kmem_caches consume less than 2GB > memory. > > After our in-depth analysis, the memory consumption of kmalloc-32 slab > cache is the cause of list_lru_one allocation. > > crash> p memcg_nr_cache_ids > memcg_nr_cache_ids = $2 = 24574 > > memcg_nr_cache_ids is very large and memory consumption of each list_lru > can be calculated with the following formula. > > num_numa_node * memcg_nr_cache_ids * 32 (kmalloc-32) > > There are 4 numa nodes in our system, so each list_lru consumes ~3MB. > > crash> list super_blocks | wc -l > 952 > > Every mount will register 2 list lrus, one is for inode, another is for > dentry. There are 952 super_blocks. So the total memory is 952 * 2 * 3 > MB (~5.6GB). But now the number of memory cgroups is less than 500. So I > guess more than 12286 memory cgroups have been created on this machine (I > do not know why there are so many cgroups, it may be a user's bug or > the user really want to do that). Because memcg_nr_cache_ids has not been > reduced to a suitable value. It leads to waste a lot of memory. If we want > to reduce memcg_nr_cache_ids, we have to *reboot* the server. This is not > what we want. > > In order to reduce memcg_nr_cache_ids, I had posted a patchset [1] to do > this. But this did not fundamentally solve the problem. > > We currently allocate scope for every memcg to be able to tracked on every > superblock instantiated in the system, regardless of whether that superblock > is even accessible to that memcg. > > These huge memcg counts come from container hosts where memcgs are confined > to just a small subset of the total number of superblocks that instantiated > at any given point in time. > > For these systems with huge container counts, list_lru does not need the > capability of tracking every memcg on every superblock. > > What it comes down to is that the list_lru is only needed for a given memcg > if that memcg is instatiating and freeing objects on a given list_lru. > > As Dave said, "Which makes me think we should be moving more towards 'add the > memcg to the list_lru at the first insert' model rather than 'instantiate > all at memcg init time just in case'." > > This patchset aims to optimize the list lru memory consumption from different > aspects. > > Patch 1-6 are code simplification. > Patch 7 converts the array from per-memcg per-node to per-memcg > Patch 8 introduces kmem_cache_alloc_lru() > Patch 9 introduces alloc_inode_sb() > Patch 10-66 convert all filesystems to alloc_inode_sb() respectively. There is now days also ntfs3. If you do not plan to convert this please CC me atleast so that I can do it when these lands. Argillander > Patch 70 let list_lru allocation dynamically. > Patch 72 use xarray to optimize per memcg pointer array size. > Patch 73-76 is code simplification. > > I had done a easy test to show the optimization. I create 10k memory cgroups > and mount 10k filesystems in the systems. We use free command to show how many > memory does the systems comsumes after this operation (There are 2 numa nodes > in the system). > > +-----------------------+------------------------+ > | condition | memory consumption | > +-----------------------+------------------------+ > | without this patchset | 24464 MB | > +-----------------------+------------------------+ > | after patch 7 | 21957 MB | <--------+ > +-----------------------+------------------------+ | > | after patch 70 | 6895 MB | | > +-----------------------+------------------------+ | > | after patch 72 | 4367 MB | | > +-----------------------+------------------------+ | > | > The more the number of nodes, the more obvious the effect---+ > > BTW, there was a recent discussion [2] on the same issue. > > [1] https://lore.kernel.org/linux-fsdevel/20210428094949.43579-1-songmuchun@bytedance.com/ > [2] https://lore.kernel.org/linux-fsdevel/20210405054848.GA1077931@in.ibm.com/ > > This series not only optimizes the memory usage of list_lru but also > simplifies the code. > > Changelog in v3: > - Fix mixing advanced and normal XArray concepts (Thanks to Matthew). > - Split one patch into per-filesystem patches. > > Changelog in v2: > - Update Documentation/filesystems/porting.rst suggested by Dave. > - Add a comment above alloc_inode_sb() suggested by Dave. > - Rework some patch's commit log. > - Add patch 18-21. > > Thanks Dave. > > Muchun Song (76): > mm: list_lru: fix the return value of list_lru_count_one() > mm: memcontrol: remove kmemcg_id reparenting > mm: memcontrol: remove the kmem states > mm: memcontrol: move memcg_online_kmem() to mem_cgroup_css_online() > mm: list_lru: remove holding lru lock > mm: list_lru: only add memcg-aware lrus to the global lru list > mm: list_lru: optimize memory consumption of arrays > mm: introduce kmem_cache_alloc_lru > fs: introduce alloc_inode_sb() to allocate filesystems specific inode > dax: allocate inode by using alloc_inode_sb() > 9p: allocate inode by using alloc_inode_sb() > adfs: allocate inode by using alloc_inode_sb() > affs: allocate inode by using alloc_inode_sb() > afs: allocate inode by using alloc_inode_sb() > befs: allocate inode by using alloc_inode_sb() > bfs: allocate inode by using alloc_inode_sb() > block: allocate inode by using alloc_inode_sb() > btrfs: allocate inode by using alloc_inode_sb() > ceph: allocate inode by using alloc_inode_sb() > cifs: allocate inode by using alloc_inode_sb() > coda: allocate inode by using alloc_inode_sb() > ecryptfs: allocate inode by using alloc_inode_sb() > efs: allocate inode by using alloc_inode_sb() > erofs: allocate inode by using alloc_inode_sb() > exfat: allocate inode by using alloc_inode_sb() > ext2: allocate inode by using alloc_inode_sb() > ext4: allocate inode by using alloc_inode_sb() > fat: allocate inode by using alloc_inode_sb() > freevxfs: allocate inode by using alloc_inode_sb() > fuse: allocate inode by using alloc_inode_sb() > gfs2: allocate inode by using alloc_inode_sb() > hfs: allocate inode by using alloc_inode_sb() > hfsplus: allocate inode by using alloc_inode_sb() > hostfs: allocate inode by using alloc_inode_sb() > hpfs: allocate inode by using alloc_inode_sb() > hugetlbfs: allocate inode by using alloc_inode_sb() > isofs: allocate inode by using alloc_inode_sb() > jffs2: allocate inode by using alloc_inode_sb() > jfs: allocate inode by using alloc_inode_sb() > minix: allocate inode by using alloc_inode_sb() > nfs: allocate inode by using alloc_inode_sb() > nilfs2: allocate inode by using alloc_inode_sb() > ntfs: allocate inode by using alloc_inode_sb() > ocfs2: allocate inode by using alloc_inode_sb() > openpromfs: allocate inode by using alloc_inode_sb() > orangefs: allocate inode by using alloc_inode_sb() > overlayfs: allocate inode by using alloc_inode_sb() > proc: allocate inode by using alloc_inode_sb() > qnx4: allocate inode by using alloc_inode_sb() > qnx6: allocate inode by using alloc_inode_sb() > reiserfs: allocate inode by using alloc_inode_sb() > romfs: allocate inode by using alloc_inode_sb() > squashfs: allocate inode by using alloc_inode_sb() > sysv: allocate inode by using alloc_inode_sb() > ubifs: allocate inode by using alloc_inode_sb() > udf: allocate inode by using alloc_inode_sb() > ufs: allocate inode by using alloc_inode_sb() > vboxsf: allocate inode by using alloc_inode_sb() > xfs: allocate inode by using alloc_inode_sb() > zonefs: allocate inode by using alloc_inode_sb() > ipc: allocate inode by using alloc_inode_sb() > shmem: allocate inode by using alloc_inode_sb() > net: allocate inode by using alloc_inode_sb() > rpc: allocate inode by using alloc_inode_sb() > f2fs: allocate inode by using alloc_inode_sb() > nfs42: use a specific kmem_cache to allocate nfs4_xattr_entry > mm: dcache: use kmem_cache_alloc_lru() to allocate dentry > xarray: use kmem_cache_alloc_lru to allocate xa_node > mm: workingset: use xas_set_lru() to pass shadow_nodes > mm: list_lru: allocate list_lru_one only when needed > mm: list_lru: rename memcg_drain_all_list_lrus to > memcg_reparent_list_lrus > mm: list_lru: replace linear array with xarray > mm: memcontrol: reuse memory cgroup ID for kmem ID > mm: memcontrol: fix cannot alloc the maximum memcg ID > mm: list_lru: rename list_lru_per_memcg to list_lru_memcg > mm: memcontrol: rename memcg_cache_id to memcg_kmem_id > > Documentation/filesystems/porting.rst | 5 + > drivers/dax/super.c | 2 +- > fs/9p/vfs_inode.c | 2 +- > fs/adfs/super.c | 2 +- > fs/affs/super.c | 2 +- > fs/afs/super.c | 2 +- > fs/befs/linuxvfs.c | 2 +- > fs/bfs/inode.c | 2 +- > fs/block_dev.c | 2 +- > fs/btrfs/inode.c | 2 +- > fs/ceph/inode.c | 2 +- > fs/cifs/cifsfs.c | 2 +- > fs/coda/inode.c | 2 +- > fs/dcache.c | 3 +- > fs/ecryptfs/super.c | 2 +- > fs/efs/super.c | 2 +- > fs/erofs/super.c | 2 +- > fs/exfat/super.c | 2 +- > fs/ext2/super.c | 2 +- > fs/ext4/super.c | 2 +- > fs/f2fs/super.c | 8 +- > fs/fat/inode.c | 2 +- > fs/freevxfs/vxfs_super.c | 2 +- > fs/fuse/inode.c | 2 +- > fs/gfs2/super.c | 2 +- > fs/hfs/super.c | 2 +- > fs/hfsplus/super.c | 2 +- > fs/hostfs/hostfs_kern.c | 2 +- > fs/hpfs/super.c | 2 +- > fs/hugetlbfs/inode.c | 2 +- > fs/inode.c | 2 +- > fs/isofs/inode.c | 2 +- > fs/jffs2/super.c | 2 +- > fs/jfs/super.c | 2 +- > fs/minix/inode.c | 2 +- > fs/nfs/inode.c | 2 +- > fs/nfs/nfs42xattr.c | 95 ++++--- > fs/nilfs2/super.c | 2 +- > fs/ntfs/inode.c | 2 +- > fs/ocfs2/dlmfs/dlmfs.c | 2 +- > fs/ocfs2/super.c | 2 +- > fs/openpromfs/inode.c | 2 +- > fs/orangefs/super.c | 2 +- > fs/overlayfs/super.c | 2 +- > fs/proc/inode.c | 2 +- > fs/qnx4/inode.c | 2 +- > fs/qnx6/inode.c | 2 +- > fs/reiserfs/super.c | 2 +- > fs/romfs/super.c | 2 +- > fs/squashfs/super.c | 2 +- > fs/sysv/inode.c | 2 +- > fs/ubifs/super.c | 2 +- > fs/udf/super.c | 2 +- > fs/ufs/super.c | 2 +- > fs/vboxsf/super.c | 2 +- > fs/xfs/xfs_icache.c | 2 +- > fs/zonefs/super.c | 2 +- > include/linux/fs.h | 11 + > include/linux/list_lru.h | 16 +- > include/linux/memcontrol.h | 49 ++-- > include/linux/slab.h | 3 + > include/linux/swap.h | 5 +- > include/linux/xarray.h | 9 +- > ipc/mqueue.c | 2 +- > lib/xarray.c | 10 +- > mm/list_lru.c | 472 ++++++++++++++++------------------ > mm/memcontrol.c | 190 ++------------ > mm/shmem.c | 2 +- > mm/slab.c | 39 ++- > mm/slab.h | 17 +- > mm/slob.c | 6 + > mm/slub.c | 42 ++- > mm/workingset.c | 2 +- > net/socket.c | 2 +- > net/sunrpc/rpc_pipe.c | 2 +- > 75 files changed, 498 insertions(+), 598 deletions(-) > > -- > 2.11.0 >
On Sat, Sep 18, 2021 at 2:56 PM Kari Argillander <kari.argillander@gmail.com> wrote: > > On Tue, Sep 14, 2021 at 03:28:22PM +0800, Muchun Song wrote: > > We introduced alloc_inode_sb() in previous version 2, which sets up the > > inode reclaim context properly, to allocate filesystems specific inode. > > So we have to convert to new API for all filesystems, which is done in > > one patch. Some filesystems are easy to convert (just replace > > kmem_cache_alloc() to alloc_inode_sb()), while other filesystems need to > > do more work. In order to make it easy for maintainers of different > > filesystems to review their own maintained part, I split the patch into > > patches which are per-filesystem in this version. I am not sure if this > > is a good idea, because there is going to be more commits. > > > > In our server, we found a suspected memory leak problem. The kmalloc-32 > > consumes more than 6GB of memory. Other kmem_caches consume less than 2GB > > memory. > > > > After our in-depth analysis, the memory consumption of kmalloc-32 slab > > cache is the cause of list_lru_one allocation. > > > > crash> p memcg_nr_cache_ids > > memcg_nr_cache_ids = $2 = 24574 > > > > memcg_nr_cache_ids is very large and memory consumption of each list_lru > > can be calculated with the following formula. > > > > num_numa_node * memcg_nr_cache_ids * 32 (kmalloc-32) > > > > There are 4 numa nodes in our system, so each list_lru consumes ~3MB. > > > > crash> list super_blocks | wc -l > > 952 > > > > Every mount will register 2 list lrus, one is for inode, another is for > > dentry. There are 952 super_blocks. So the total memory is 952 * 2 * 3 > > MB (~5.6GB). But now the number of memory cgroups is less than 500. So I > > guess more than 12286 memory cgroups have been created on this machine (I > > do not know why there are so many cgroups, it may be a user's bug or > > the user really want to do that). Because memcg_nr_cache_ids has not been > > reduced to a suitable value. It leads to waste a lot of memory. If we want > > to reduce memcg_nr_cache_ids, we have to *reboot* the server. This is not > > what we want. > > > > In order to reduce memcg_nr_cache_ids, I had posted a patchset [1] to do > > this. But this did not fundamentally solve the problem. > > > > We currently allocate scope for every memcg to be able to tracked on every > > superblock instantiated in the system, regardless of whether that superblock > > is even accessible to that memcg. > > > > These huge memcg counts come from container hosts where memcgs are confined > > to just a small subset of the total number of superblocks that instantiated > > at any given point in time. > > > > For these systems with huge container counts, list_lru does not need the > > capability of tracking every memcg on every superblock. > > > > What it comes down to is that the list_lru is only needed for a given memcg > > if that memcg is instatiating and freeing objects on a given list_lru. > > > > As Dave said, "Which makes me think we should be moving more towards 'add the > > memcg to the list_lru at the first insert' model rather than 'instantiate > > all at memcg init time just in case'." > > > > This patchset aims to optimize the list lru memory consumption from different > > aspects. > > > > Patch 1-6 are code simplification. > > Patch 7 converts the array from per-memcg per-node to per-memcg > > Patch 8 introduces kmem_cache_alloc_lru() > > Patch 9 introduces alloc_inode_sb() > > Patch 10-66 convert all filesystems to alloc_inode_sb() respectively. > > There is now days also ntfs3. If you do not plan to convert this please > CC me atleast so that I can do it when these lands. > > Argillander > Wow, a new filesystem. I didn't notice it before. I'll cover it in the next version and Cc you if you can do a review. Thanks for your reminder.