| Message ID | 171956951930.1897969.8709279863947931285.stgit@firesoul (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [V5] cgroup/rstat: Avoid thundering herd problem by kswapd across NUMA nodes | expand |
(production results below) On 28/06/2024 12.12, Jesper Dangaard Brouer wrote: > Avoid lock contention on the global cgroup rstat lock caused by kswapd > starting on all NUMA nodes simultaneously. At Cloudflare, we observed > massive issues due to kswapd and the specific mem_cgroup_flush_stats() > call inlined in shrink_node, which takes the rstat lock. > > On our 12 NUMA node machines, each with a kswapd kthread per NUMA node, > we noted severe lock contention on the rstat lock. This contention > causes 12 CPUs to waste cycles spinning every time kswapd runs. > Fleet-wide stats (/proc/N/schedstat) for kthreads revealed that we are > burning an average of 20,000 CPU cores fleet-wide on kswapd, primarily > due to spinning on the rstat lock. > > Help reviewers follow code: __alloc_pages_slowpath calls wake_all_kswapds > causing all kswapdN threads to wake up simultaneously. The kswapd thread > invokes shrink_node (via balance_pgdat) triggering the cgroup rstat flush > operation as part of its work. This results in kernel self-induced rstat > lock contention by waking up all kswapd threads simultaneously. Leveraging > this detail: balance_pgdat() have NULL value in target_mem_cgroup, this > cause mem_cgroup_flush_stats() to do flush with root_mem_cgroup. > > To avoid this kind of thundering herd problem, kernel previously had a > "stats_flush_ongoing" concept, but this was removed as part of commit > 7d7ef0a4686a ("mm: memcg: restore subtree stats flushing"). This patch > reintroduce and generalized the concept to apply to all users of cgroup > rstat, not just memcg. > > If there is an ongoing rstat flush, and current cgroup is a descendant, > then it is unnecessary to do the flush. For callers to still see updated > stats, wait for ongoing flusher to complete before returning, but add > timeout as stats are already inaccurate given updaters keeps running. > > Fixes: 7d7ef0a4686a ("mm: memcg: restore subtree stats flushing"). > Signed-off-by: Jesper Dangaard Brouer <hawk@kernel.org> > --- > V5: > Dropped trylock helper > > V4: https://lore.kernel.org/all/171952312320.1810550.13209360603489797077.stgit@firesoul/ > V3: https://lore.kernel.org/all/171943668946.1638606.1320095353103578332.stgit@firesoul/ > V2: https://lore.kernel.org/all/171923011608.1500238.3591002573732683639.stgit@firesoul/ > V1: https://lore.kernel.org/all/171898037079.1222367.13467317484793748519.stgit@firesoul/ > RFC: https://lore.kernel.org/all/171895533185.1084853.3033751561302228252.stgit@firesoul/ > > include/linux/cgroup-defs.h | 2 ++ > kernel/cgroup/rstat.c | 57 +++++++++++++++++++++++++++++++++++++++---- > 2 files changed, 54 insertions(+), 5 deletions(-) > > diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h > index b36690ca0d3f..a33b37514c29 100644 > --- a/include/linux/cgroup-defs.h > +++ b/include/linux/cgroup-defs.h > @@ -548,6 +548,8 @@ struct cgroup { > #ifdef CONFIG_BPF_SYSCALL > struct bpf_local_storage __rcu *bpf_cgrp_storage; > #endif > + /* completion queue for cgrp_rstat_ongoing_flusher */ > + struct completion flush_done; > > /* All ancestors including self */ > struct cgroup *ancestors[]; > diff --git a/kernel/cgroup/rstat.c b/kernel/cgroup/rstat.c > index fb8b49437573..e9d3e2aff698 100644 > --- a/kernel/cgroup/rstat.c > +++ b/kernel/cgroup/rstat.c > @@ -2,6 +2,7 @@ > #include "cgroup-internal.h" > > #include <linux/sched/cputime.h> > +#include <linux/completion.h> > > #include <linux/bpf.h> > #include <linux/btf.h> > @@ -11,6 +12,7 @@ > > static DEFINE_SPINLOCK(cgroup_rstat_lock); > static DEFINE_PER_CPU(raw_spinlock_t, cgroup_rstat_cpu_lock); > +static struct cgroup *cgrp_rstat_ongoing_flusher = NULL; > > static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu); > > @@ -299,6 +301,44 @@ static inline void __cgroup_rstat_unlock(struct cgroup *cgrp, int cpu_in_loop) > spin_unlock_irq(&cgroup_rstat_lock); > } > > +#define MAX_WAIT msecs_to_jiffies(100) > +/* Trylock helper that also checks for on ongoing flusher */ > +static bool cgroup_rstat_trylock_flusher(struct cgroup *cgrp) > +{ > + struct cgroup *cgrp_ongoing; > + > + /* Check if ongoing flusher is already taking care of this, if > + * we are a descendant skip work, but wait for ongoing flusher > + * to complete work. > + */ > + cgrp_ongoing = READ_ONCE(cgrp_rstat_ongoing_flusher); > + if (cgrp_ongoing && cgroup_is_descendant(cgrp, cgrp_ongoing)) { > + wait_for_completion_interruptible_timeout( > + &cgrp_ongoing->flush_done, MAX_WAIT); > + /* TODO: Add tracepoint here */ > + return false; > + } > + > + __cgroup_rstat_lock(cgrp, -1); > + /* Obtained lock, record this cgrp as the ongoing flusher */ > + if (!READ_ONCE(cgrp_rstat_ongoing_flusher)) { > + reinit_completion(&cgrp->flush_done); > + WRITE_ONCE(cgrp_rstat_ongoing_flusher, cgrp); > + } > + > + return true; /* locked */ > +} > + This scheme doesn't fully stop the kswapd thundering herd problem. Below bpftrace script look at cgroup_rstat_trylock_flusher() return values. If false (0) is returned then we avoided a flush operation and locking. [Data from production- AMD with 12 NUMA nodes] bpftrace -e 'kretfunc:cgroup_rstat_trylock_flusher {@cnt = hist(retval)} interval:s:10 {time("%H:%M:%S "); print(@cnt); clear(@cnt)}' Attaching 2 probes... 11:42:28 @cnt: [0] 467 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 203 |@@@@@@@@@@@@@@@@@@@@@@ | 11:42:38 @cnt: [0] 320 |@@@@@@@@@@@@@ | [1] 1217 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| 11:42:48 @cnt: [0] 256 |@@@@@@@@@@@@@ | [1] 969 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| Above data shows that, the success rate (for avoiding concurrent flushing) isn't as high as we would expect and had hoped for. I'm not sure what is the reason for this? I can only spegulate. (A) Pergaps the READ_ONCE() outside the lock isn't sufficient. This would imply that wake_all_kswapds() wakes up at exactly same time. (B) Perhaps wake_all_kswapds() is not exact and there isn't any contention and ongoing flusher. Unlikely, as we can measure contention (see below for more data), and lockhold time is between 64 usec to 256 usec (with outliers around 32 ms). (C) Sub-tree ongoing flusher could be blocking us. This, can happen and I caught come cases of this happening when cadvisor collects data. But for cases where it is only kswapdN running this also happens (see below data with 10 sec period with kswapd and periodic flusher). Idea are welcome for what I should try first? > +static void cgroup_rstat_unlock_flusher(struct cgroup *cgrp) > +{ > + /* Detect if we are the ongoing flusher */ > + if (cgrp == READ_ONCE(cgrp_rstat_ongoing_flusher)) { > + WRITE_ONCE(cgrp_rstat_ongoing_flusher, NULL); > + complete_all(&cgrp->flush_done); > + } > + __cgroup_rstat_unlock(cgrp, -1); > +} > + > /* see cgroup_rstat_flush() */ > static void cgroup_rstat_flush_locked(struct cgroup *cgrp) > __releases(&cgroup_rstat_lock) __acquires(&cgroup_rstat_lock) > @@ -350,9 +390,11 @@ __bpf_kfunc void cgroup_rstat_flush(struct cgroup *cgrp) > { > might_sleep(); > > - __cgroup_rstat_lock(cgrp, -1); > + if (!cgroup_rstat_trylock_flusher(cgrp)) > + return; > + > cgroup_rstat_flush_locked(cgrp); > - __cgroup_rstat_unlock(cgrp, -1); > + cgroup_rstat_unlock_flusher(cgrp); > } > > /** > @@ -368,8 +410,11 @@ void cgroup_rstat_flush_hold(struct cgroup *cgrp) > __acquires(&cgroup_rstat_lock) > { > might_sleep(); > - __cgroup_rstat_lock(cgrp, -1); > - cgroup_rstat_flush_locked(cgrp); > + > + if (cgroup_rstat_trylock_flusher(cgrp)) > + cgroup_rstat_flush_locked(cgrp); > + else > + __cgroup_rstat_lock(cgrp, -1); > } > > /** > @@ -379,7 +424,7 @@ void cgroup_rstat_flush_hold(struct cgroup *cgrp) > void cgroup_rstat_flush_release(struct cgroup *cgrp) > __releases(&cgroup_rstat_lock) > { > - __cgroup_rstat_unlock(cgrp, -1); > + cgroup_rstat_unlock_flusher(cgrp); > } > > int cgroup_rstat_init(struct cgroup *cgrp) > @@ -401,6 +446,8 @@ int cgroup_rstat_init(struct cgroup *cgrp) > u64_stats_init(&rstatc->bsync); > } > > + init_completion(&cgrp->flush_done); > + > return 0; > } > > > Data including 'comm' bpftrace -e 'kretfunc:cgroup_rstat_trylock_flusher {@total=hist(retval);@ret[comm] = hist(retval)} interval:s:10 {time("%H:%M:%S "); print(@ret); print(@total); clear(@ret)}' Attaching 2 probes... 12:13:21 @ret[kworker/u388:15]: [1] 1 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| @ret[kworker/u389:20]: [1] 2 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| @ret[kworker/u389:3]: [1] 2 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| @ret[kswapd3]: [0] 10 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 1 |@@@@@ | @ret[kswapd11]: [0] 10 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 2 |@@@@@@@@@@ | @ret[kswapd2]: [0] 12 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 1 |@@@@ | @ret[kswapd10]: [0] 10 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 5 |@@@@@@@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd9]: [0] 13 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 2 |@@@@@@@@ | @ret[kswapd0]: [0] 11 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 5 |@@@@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd6]: [0] 9 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 8 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd7]: [0] 13 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 5 |@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd4]: [0] 12 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 7 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd5]: [0] 13 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 6 |@@@@@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd1]: [0] 16 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 9 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | @ret[kswapd8]: [0] 14 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 14 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| @total: [0] 135 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [1] 70 |@@@@@@@@@@@@@@@@@@@@@@@@@@ | How long time is the lock locked (@locked_ns) and how long does lock contenders wait to obtain lock (@wait_ns). bpftrace -e ' tracepoint:cgroup:cgroup_rstat_lock_contended {@start[tid]=nsecs} tracepoint:cgroup:cgroup_rstat_locked { @locked[tid]=nsecs; if (args->contended) { @wait_ns=hist(nsecs-@start[tid]); delete(@start[tid]);}} tracepoint:cgroup:cgroup_rstat_unlock { @locked_ns=hist(nsecs-@locked[tid]); delete(@locked[tid]);} interval:s:1 {time("%H:%M:%S "); print(@wait_ns);print(@locked_ns); }' @locked_ns: [256, 512) 94 | | [512, 1K) 5870 |@@@@@@@@@@@@@@@@@@@@@@@@@@ | [1K, 2K) 5353 |@@@@@@@@@@@@@@@@@@@@@@@@ | [2K, 4K) 439 |@ | [4K, 8K) 172 | | [8K, 16K) 6344 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | [16K, 32K) 6954 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | [32K, 64K) 2700 |@@@@@@@@@@@@ | [64K, 128K) 11451 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [128K, 256K) 8362 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | [256K, 512K) 1167 |@@@@@ | [512K, 1M) 114 | | [1M, 2M) 32 | | [2M, 4M) 25 | | [4M, 8M) 55 | | [8M, 16M) 258 |@ | [16M, 32M) 221 |@ | [32M, 64M) 24 | | @wait_ns: [512, 1K) 30 | | [1K, 2K) 2124 |@@@@@@@ | [2K, 4K) 3319 |@@@@@@@@@@@ | [4K, 8K) 1595 |@@@@@ | [8K, 16K) 3288 |@@@@@@@@@@@ | [16K, 32K) 5349 |@@@@@@@@@@@@@@@@@@ | [32K, 64K) 5069 |@@@@@@@@@@@@@@@@@ | [64K, 128K) 15451 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@| [128K, 256K) 9420 |@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ | [256K, 512K) 707 |@@ | [512K, 1M) 2 | | --Jesper
diff --git a/include/linux/cgroup-defs.h b/include/linux/cgroup-defs.h index b36690ca0d3f..a33b37514c29 100644 --- a/include/linux/cgroup-defs.h +++ b/include/linux/cgroup-defs.h @@ -548,6 +548,8 @@ struct cgroup { #ifdef CONFIG_BPF_SYSCALL struct bpf_local_storage __rcu *bpf_cgrp_storage; #endif + /* completion queue for cgrp_rstat_ongoing_flusher */ + struct completion flush_done; /* All ancestors including self */ struct cgroup *ancestors[]; diff --git a/kernel/cgroup/rstat.c b/kernel/cgroup/rstat.c index fb8b49437573..e9d3e2aff698 100644 --- a/kernel/cgroup/rstat.c +++ b/kernel/cgroup/rstat.c @@ -2,6 +2,7 @@ #include "cgroup-internal.h" #include <linux/sched/cputime.h> +#include <linux/completion.h> #include <linux/bpf.h> #include <linux/btf.h> @@ -11,6 +12,7 @@ static DEFINE_SPINLOCK(cgroup_rstat_lock); static DEFINE_PER_CPU(raw_spinlock_t, cgroup_rstat_cpu_lock); +static struct cgroup *cgrp_rstat_ongoing_flusher = NULL; static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu); @@ -299,6 +301,44 @@ static inline void __cgroup_rstat_unlock(struct cgroup *cgrp, int cpu_in_loop) spin_unlock_irq(&cgroup_rstat_lock); } +#define MAX_WAIT msecs_to_jiffies(100) +/* Trylock helper that also checks for on ongoing flusher */ +static bool cgroup_rstat_trylock_flusher(struct cgroup *cgrp) +{ + struct cgroup *cgrp_ongoing; + + /* Check if ongoing flusher is already taking care of this, if + * we are a descendant skip work, but wait for ongoing flusher + * to complete work. + */ + cgrp_ongoing = READ_ONCE(cgrp_rstat_ongoing_flusher); + if (cgrp_ongoing && cgroup_is_descendant(cgrp, cgrp_ongoing)) { + wait_for_completion_interruptible_timeout( + &cgrp_ongoing->flush_done, MAX_WAIT); + /* TODO: Add tracepoint here */ + return false; + } + + __cgroup_rstat_lock(cgrp, -1); + /* Obtained lock, record this cgrp as the ongoing flusher */ + if (!READ_ONCE(cgrp_rstat_ongoing_flusher)) { + reinit_completion(&cgrp->flush_done); + WRITE_ONCE(cgrp_rstat_ongoing_flusher, cgrp); + } + + return true; /* locked */ +} + +static void cgroup_rstat_unlock_flusher(struct cgroup *cgrp) +{ + /* Detect if we are the ongoing flusher */ + if (cgrp == READ_ONCE(cgrp_rstat_ongoing_flusher)) { + WRITE_ONCE(cgrp_rstat_ongoing_flusher, NULL); + complete_all(&cgrp->flush_done); + } + __cgroup_rstat_unlock(cgrp, -1); +} + /* see cgroup_rstat_flush() */ static void cgroup_rstat_flush_locked(struct cgroup *cgrp) __releases(&cgroup_rstat_lock) __acquires(&cgroup_rstat_lock) @@ -350,9 +390,11 @@ __bpf_kfunc void cgroup_rstat_flush(struct cgroup *cgrp) { might_sleep(); - __cgroup_rstat_lock(cgrp, -1); + if (!cgroup_rstat_trylock_flusher(cgrp)) + return; + cgroup_rstat_flush_locked(cgrp); - __cgroup_rstat_unlock(cgrp, -1); + cgroup_rstat_unlock_flusher(cgrp); } /** @@ -368,8 +410,11 @@ void cgroup_rstat_flush_hold(struct cgroup *cgrp) __acquires(&cgroup_rstat_lock) { might_sleep(); - __cgroup_rstat_lock(cgrp, -1); - cgroup_rstat_flush_locked(cgrp); + + if (cgroup_rstat_trylock_flusher(cgrp)) + cgroup_rstat_flush_locked(cgrp); + else + __cgroup_rstat_lock(cgrp, -1); } /** @@ -379,7 +424,7 @@ void cgroup_rstat_flush_hold(struct cgroup *cgrp) void cgroup_rstat_flush_release(struct cgroup *cgrp) __releases(&cgroup_rstat_lock) { - __cgroup_rstat_unlock(cgrp, -1); + cgroup_rstat_unlock_flusher(cgrp); } int cgroup_rstat_init(struct cgroup *cgrp) @@ -401,6 +446,8 @@ int cgroup_rstat_init(struct cgroup *cgrp) u64_stats_init(&rstatc->bsync); } + init_completion(&cgrp->flush_done); + return 0; }
Avoid lock contention on the global cgroup rstat lock caused by kswapd starting on all NUMA nodes simultaneously. At Cloudflare, we observed massive issues due to kswapd and the specific mem_cgroup_flush_stats() call inlined in shrink_node, which takes the rstat lock. On our 12 NUMA node machines, each with a kswapd kthread per NUMA node, we noted severe lock contention on the rstat lock. This contention causes 12 CPUs to waste cycles spinning every time kswapd runs. Fleet-wide stats (/proc/N/schedstat) for kthreads revealed that we are burning an average of 20,000 CPU cores fleet-wide on kswapd, primarily due to spinning on the rstat lock. Help reviewers follow code: __alloc_pages_slowpath calls wake_all_kswapds causing all kswapdN threads to wake up simultaneously. The kswapd thread invokes shrink_node (via balance_pgdat) triggering the cgroup rstat flush operation as part of its work. This results in kernel self-induced rstat lock contention by waking up all kswapd threads simultaneously. Leveraging this detail: balance_pgdat() have NULL value in target_mem_cgroup, this cause mem_cgroup_flush_stats() to do flush with root_mem_cgroup. To avoid this kind of thundering herd problem, kernel previously had a "stats_flush_ongoing" concept, but this was removed as part of commit 7d7ef0a4686a ("mm: memcg: restore subtree stats flushing"). This patch reintroduce and generalized the concept to apply to all users of cgroup rstat, not just memcg. If there is an ongoing rstat flush, and current cgroup is a descendant, then it is unnecessary to do the flush. For callers to still see updated stats, wait for ongoing flusher to complete before returning, but add timeout as stats are already inaccurate given updaters keeps running. Fixes: 7d7ef0a4686a ("mm: memcg: restore subtree stats flushing"). Signed-off-by: Jesper Dangaard Brouer <hawk@kernel.org> --- V5: Dropped trylock helper V4: https://lore.kernel.org/all/171952312320.1810550.13209360603489797077.stgit@firesoul/ V3: https://lore.kernel.org/all/171943668946.1638606.1320095353103578332.stgit@firesoul/ V2: https://lore.kernel.org/all/171923011608.1500238.3591002573732683639.stgit@firesoul/ V1: https://lore.kernel.org/all/171898037079.1222367.13467317484793748519.stgit@firesoul/ RFC: https://lore.kernel.org/all/171895533185.1084853.3033751561302228252.stgit@firesoul/ include/linux/cgroup-defs.h | 2 ++ kernel/cgroup/rstat.c | 57 +++++++++++++++++++++++++++++++++++++++---- 2 files changed, 54 insertions(+), 5 deletions(-)