From 9783aa9917f8ae24759e67bf882f1aba32fe4ea1 Mon Sep 17 00:00:00 2001 From: Chris Down Date: Sun, 6 Oct 2019 17:58:32 -0700 Subject: [PATCH] mm, memcg: proportional memory.{low,min} reclaim cgroup v2 introduces two memory protection thresholds: memory.low (best-effort) and memory.min (hard protection). While they generally do what they say on the tin, there is a limitation in their implementation that makes them difficult to use effectively: that cliff behaviour often manifests when they become eligible for reclaim. This patch implements more intuitive and usable behaviour, where we gradually mount more reclaim pressure as cgroups further and further exceed their protection thresholds. This cliff edge behaviour happens because we only choose whether or not to reclaim based on whether the memcg is within its protection limits (see the use of mem_cgroup_protected in shrink_node), but we don't vary our reclaim behaviour based on this information. Imagine the following timeline, with the numbers the lruvec size in this zone: 1. memory.low=1000000, memory.current=999999. 0 pages may be scanned. 2. memory.low=1000000, memory.current=1000000. 0 pages may be scanned. 3. memory.low=1000000, memory.current=1000001. 1000001* pages may be scanned. (?!) * Of course, we won't usually scan all available pages in the zone even without this patch because of scan control priority, over-reclaim protection, etc. However, as shown by the tests at the end, these techniques don't sufficiently throttle such an extreme change in input, so cliff-like behaviour isn't really averted by their existence alone. Here's an example of how this plays out in practice. At Facebook, we are trying to protect various workloads from "system" software, like configuration management tools, metric collectors, etc (see this[0] case study). In order to find a suitable memory.low value, we start by determining the expected memory range within which the workload will be comfortable operating. This isn't an exact science -- memory usage deemed "comfortable" will vary over time due to user behaviour, differences in composition of work, etc, etc. As such we need to ballpark memory.low, but doing this is currently problematic: 1. If we end up setting it too low for the workload, it won't have *any* effect (see discussion above). The group will receive the full weight of reclaim and won't have any priority while competing with the less important system software, as if we had no memory.low configured at all. 2. Because of this behaviour, we end up erring on the side of setting it too high, such that the comfort range is reliably covered. However, protected memory is completely unavailable to the rest of the system, so we might cause undue memory and IO pressure there when we *know* we have some elasticity in the workload. 3. Even if we get the value totally right, smack in the middle of the comfort zone, we get extreme jumps between no pressure and full pressure that cause unpredictable pressure spikes in the workload due to the current binary reclaim behaviour. With this patch, we can set it to our ballpark estimation without too much worry. Any undesirable behaviour, such as too much or too little reclaim pressure on the workload or system will be proportional to how far our estimation is off. This means we can set memory.low much more conservatively and thus waste less resources *without* the risk of the workload falling off a cliff if we overshoot. As a more abstract technical description, this unintuitive behaviour results in having to give high-priority workloads a large protection buffer on top of their expected usage to function reliably, as otherwise we have abrupt periods of dramatically increased memory pressure which hamper performance. Having to set these thresholds so high wastes resources and generally works against the principle of work conservation. In addition, having proportional memory reclaim behaviour has other benefits. Most notably, before this patch it's basically mandatory to set memory.low to a higher than desirable value because otherwise as soon as you exceed memory.low, all protection is lost, and all pages are eligible to scan again. By contrast, having a gradual ramp in reclaim pressure means that you now still get some protection when thresholds are exceeded, which means that one can now be more comfortable setting memory.low to lower values without worrying that all protection will be lost. This is important because workingset size is really hard to know exactly, especially with variable workloads, so at least getting *some* protection if your workingset size grows larger than you expect increases user confidence in setting memory.low without a huge buffer on top being needed. Thanks a lot to Johannes Weiner and Tejun Heo for their advice and assistance in thinking about how to make this work better. In testing these changes, I intended to verify that: 1. Changes in page scanning become gradual and proportional instead of binary. To test this, I experimented stepping further and further down memory.low protection on a workload that floats around 19G workingset when under memory.low protection, watching page scan rates for the workload cgroup: +------------+-----------------+--------------------+--------------+ | memory.low | test (pgscan/s) | control (pgscan/s) | % of control | +------------+-----------------+--------------------+--------------+ | 21G | 0 | 0 | N/A | | 17G | 867 | 3799 | 23% | | 12G | 1203 | 3543 | 34% | | 8G | 2534 | 3979 | 64% | | 4G | 3980 | 4147 | 96% | | 0 | 3799 | 3980 | 95% | +------------+-----------------+--------------------+--------------+ As you can see, the test kernel (with a kernel containing this patch) ramps up page scanning significantly more gradually than the control kernel (without this patch). 2. More gradual ramp up in reclaim aggression doesn't result in premature OOMs. To test this, I wrote a script that slowly increments the number of pages held by stress(1)'s --vm-keep mode until a production system entered severe overall memory contention. This script runs in a highly protected slice taking up the majority of available system memory. Watching vmstat revealed that page scanning continued essentially nominally between test and control, without causing forward reclaim progress to become arrested. [0]: https://facebookmicrosites.github.io/cgroup2/docs/overview.html#case-study-the-fbtax2-project [akpm@linux-foundation.org: reflow block comments to fit in 80 cols] [chris@chrisdown.name: handle cgroup_disable=memory when getting memcg protection] Link: http://lkml.kernel.org/r/20190201045711.GA18302@chrisdown.name Link: http://lkml.kernel.org/r/20190124014455.GA6396@chrisdown.name Signed-off-by: Chris Down Acked-by: Johannes Weiner Reviewed-by: Roman Gushchin Cc: Michal Hocko Cc: Tejun Heo Cc: Dennis Zhou Cc: Tetsuo Handa Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- Documentation/admin-guide/cgroup-v2.rst | 20 ++++-- include/linux/memcontrol.h | 20 ++++++ mm/memcontrol.c | 5 ++ mm/vmscan.c | 82 +++++++++++++++++++++++-- 4 files changed, 115 insertions(+), 12 deletions(-) diff --git a/Documentation/admin-guide/cgroup-v2.rst b/Documentation/admin-guide/cgroup-v2.rst index 0fa8c0e615c2..5361ebec3361 100644 --- a/Documentation/admin-guide/cgroup-v2.rst +++ b/Documentation/admin-guide/cgroup-v2.rst @@ -615,8 +615,8 @@ on an IO device and is an example of this type. Protections ----------- -A cgroup is protected to be allocated upto the configured amount of -the resource if the usages of all its ancestors are under their +A cgroup is protected upto the configured amount of the resource +as long as the usages of all its ancestors are under their protected levels. Protections can be hard guarantees or best effort soft boundaries. Protections can also be over-committed in which case only upto the amount available to the parent is protected among @@ -1096,7 +1096,10 @@ PAGE_SIZE multiple when read back. is within its effective min boundary, the cgroup's memory won't be reclaimed under any conditions. If there is no unprotected reclaimable memory available, OOM killer - is invoked. + is invoked. Above the effective min boundary (or + effective low boundary if it is higher), pages are reclaimed + proportionally to the overage, reducing reclaim pressure for + smaller overages. Effective min boundary is limited by memory.min values of all ancestor cgroups. If there is memory.min overcommitment @@ -1118,7 +1121,10 @@ PAGE_SIZE multiple when read back. Best-effort memory protection. If the memory usage of a cgroup is within its effective low boundary, the cgroup's memory won't be reclaimed unless memory can be reclaimed - from unprotected cgroups. + from unprotected cgroups. Above the effective low boundary (or + effective min boundary if it is higher), pages are reclaimed + proportionally to the overage, reducing reclaim pressure for + smaller overages. Effective low boundary is limited by memory.low values of all ancestor cgroups. If there is memory.low overcommitment @@ -2482,8 +2488,10 @@ system performance due to overreclaim, to the point where the feature becomes self-defeating. The memory.low boundary on the other hand is a top-down allocated -reserve. A cgroup enjoys reclaim protection when it's within its low, -which makes delegation of subtrees possible. +reserve. A cgroup enjoys reclaim protection when it's within its +effective low, which makes delegation of subtrees possible. It also +enjoys having reclaim pressure proportional to its overage when +above its effective low. The original high boundary, the hard limit, is defined as a strict limit that can not budge, even if the OOM killer has to be called. diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h index 98380779f6d5..fa9ba2edf7e0 100644 --- a/include/linux/memcontrol.h +++ b/include/linux/memcontrol.h @@ -356,6 +356,14 @@ static inline bool mem_cgroup_disabled(void) return !cgroup_subsys_enabled(memory_cgrp_subsys); } +static inline unsigned long mem_cgroup_protection(struct mem_cgroup *memcg) +{ + if (mem_cgroup_disabled()) + return 0; + + return max(READ_ONCE(memcg->memory.emin), READ_ONCE(memcg->memory.elow)); +} + enum mem_cgroup_protection mem_cgroup_protected(struct mem_cgroup *root, struct mem_cgroup *memcg); @@ -537,6 +545,8 @@ void mem_cgroup_handle_over_high(void); unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg); +unsigned long mem_cgroup_size(struct mem_cgroup *memcg); + void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p); @@ -829,6 +839,11 @@ static inline void memcg_memory_event_mm(struct mm_struct *mm, { } +static inline unsigned long mem_cgroup_protection(struct mem_cgroup *memcg) +{ + return 0; +} + static inline enum mem_cgroup_protection mem_cgroup_protected( struct mem_cgroup *root, struct mem_cgroup *memcg) { @@ -968,6 +983,11 @@ static inline unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) return 0; } +static inline unsigned long mem_cgroup_size(struct mem_cgroup *memcg) +{ + return 0; +} + static inline void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) { diff --git a/mm/memcontrol.c b/mm/memcontrol.c index c313c49074ca..bdac56009a38 100644 --- a/mm/memcontrol.c +++ b/mm/memcontrol.c @@ -1567,6 +1567,11 @@ unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) return max; } +unsigned long mem_cgroup_size(struct mem_cgroup *memcg) +{ + return page_counter_read(&memcg->memory); +} + static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, int order) { diff --git a/mm/vmscan.c b/mm/vmscan.c index e5d52d6a24af..dfefa1d99d1b 100644 --- a/mm/vmscan.c +++ b/mm/vmscan.c @@ -2459,17 +2459,80 @@ out: *lru_pages = 0; for_each_evictable_lru(lru) { int file = is_file_lru(lru); - unsigned long size; + unsigned long lruvec_size; unsigned long scan; + unsigned long protection; + + lruvec_size = lruvec_lru_size(lruvec, lru, sc->reclaim_idx); + protection = mem_cgroup_protection(memcg); + + if (protection > 0) { + /* + * Scale a cgroup's reclaim pressure by proportioning + * its current usage to its memory.low or memory.min + * setting. + * + * This is important, as otherwise scanning aggression + * becomes extremely binary -- from nothing as we + * approach the memory protection threshold, to totally + * nominal as we exceed it. This results in requiring + * setting extremely liberal protection thresholds. It + * also means we simply get no protection at all if we + * set it too low, which is not ideal. + */ + unsigned long cgroup_size = mem_cgroup_size(memcg); + unsigned long baseline = 0; + + /* + * During the reclaim first pass, we only consider + * cgroups in excess of their protection setting, but if + * that doesn't produce free pages, we come back for a + * second pass where we reclaim from all groups. + * + * To maintain fairness in both cases, the first pass + * targets groups in proportion to their overage, and + * the second pass targets groups in proportion to their + * protection utilization. + * + * So on the first pass, a group whose size is 130% of + * its protection will be targeted at 30% of its size. + * On the second pass, a group whose size is at 40% of + * its protection will be + * targeted at 40% of its size. + */ + if (!sc->memcg_low_reclaim) + baseline = lruvec_size; + scan = lruvec_size * cgroup_size / protection - baseline; + + /* + * Don't allow the scan target to exceed the lruvec + * size, which otherwise could happen if we have >200% + * overage in the normal case, or >100% overage when + * sc->memcg_low_reclaim is set. + * + * This is important because other cgroups without + * memory.low have their scan target initially set to + * their lruvec size, so allowing values >100% of the + * lruvec size here could result in penalising cgroups + * with memory.low set even *more* than their peers in + * some cases in the case of large overages. + * + * Also, minimally target SWAP_CLUSTER_MAX pages to keep + * reclaim moving forwards. + */ + scan = clamp(scan, SWAP_CLUSTER_MAX, lruvec_size); + } else { + scan = lruvec_size; + } + + scan >>= sc->priority; - size = lruvec_lru_size(lruvec, lru, sc->reclaim_idx); - scan = size >> sc->priority; /* * If the cgroup's already been deleted, make sure to * scrape out the remaining cache. */ if (!scan && !mem_cgroup_online(memcg)) - scan = min(size, SWAP_CLUSTER_MAX); + scan = min(lruvec_size, SWAP_CLUSTER_MAX); switch (scan_balance) { case SCAN_EQUAL: @@ -2489,7 +2552,7 @@ out: case SCAN_ANON: /* Scan one type exclusively */ if ((scan_balance == SCAN_FILE) != file) { - size = 0; + lruvec_size = 0; scan = 0; } break; @@ -2498,7 +2561,7 @@ out: BUG(); } - *lru_pages += size; + *lru_pages += lruvec_size; nr[lru] = scan; } } @@ -2742,6 +2805,13 @@ static bool shrink_node(pg_data_t *pgdat, struct scan_control *sc) memcg_memory_event(memcg, MEMCG_LOW); break; case MEMCG_PROT_NONE: + /* + * All protection thresholds breached. We may + * still choose to vary the scan pressure + * applied based on by how much the cgroup in + * question has exceeded its protection + * thresholds (see get_scan_count). + */ break; }