From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from lists.gnu.org (lists.gnu.org [209.51.188.17]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 28F63C433F5 for ; Fri, 1 Apr 2022 09:51:27 +0000 (UTC) Received: from localhost ([::1]:34538 helo=lists1p.gnu.org) by lists.gnu.org with esmtp (Exim 4.90_1) (envelope-from ) id 1naDwE-0000zJ-6J for qemu-devel@archiver.kernel.org; Fri, 01 Apr 2022 05:51:26 -0400 Received: from eggs.gnu.org ([209.51.188.92]:60810) by lists.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1naDgv-0000MY-U3 for qemu-devel@nongnu.org; Fri, 01 Apr 2022 05:35:37 -0400 Received: from us-smtp-delivery-124.mimecast.com ([170.10.129.124]:42066) by eggs.gnu.org with esmtps (TLS1.2:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.90_1) (envelope-from ) id 1naDgt-00018c-En for qemu-devel@nongnu.org; Fri, 01 Apr 2022 05:35:37 -0400 DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1648805734; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=NHsmVLcMXB67RkJVH556zBvO5bKxPBw/mDa94yXuc6w=; b=aXEs8UOtFhtpiVcyDD/urWl9a4q09p1lU/aU4c235ZBX09NLAi/LTgIyELUqKMXpAaZd5e kqo8MgHPHDai7qyIO5h0RDYNbXwUqfRS7LlZStowrzCM1rCjW9INl7g/0nFt1yqPnlLvVm rgE1C9e/nxq0eODlUGlXsITZLx9vOCo= Received: from mail-wr1-f71.google.com (mail-wr1-f71.google.com [209.85.221.71]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id us-mta-671-bv93ofbpNouyEdufYFKTOw-1; Fri, 01 Apr 2022 05:35:33 -0400 X-MC-Unique: bv93ofbpNouyEdufYFKTOw-1 Received: by mail-wr1-f71.google.com with SMTP id u18-20020adfa192000000b00205f0d87876so478282wru.4 for ; Fri, 01 Apr 2022 02:35:33 -0700 (PDT) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:in-reply-to :references:mime-version:content-transfer-encoding; bh=NHsmVLcMXB67RkJVH556zBvO5bKxPBw/mDa94yXuc6w=; b=Wu9vRC1+3bYyNUjsY2LC4trvYCGeAigyNchCo7nrBXSZek917x18o1M6g2OhVgUBAw Fmp+G+axGt8EsVEaauj8OdakNYPdZyEe/bE1uW50xiKBNLUhw94CC8LNjCnUeKAA4cN3 7Ajkz/aq8gnLMr4xlaO6S0XJM04KTzl9GBf6vrnb/h13Lr7je5Mx6vwUONALDR6SLyWV 5qbFT+HAayqfvYN6stMIqxfB1zGwK/sg4oWHxiWMeKCIxLDOBIDRNbwUFrV9IOPcUD2C VYF5W2n2jtO4nf1RlaOEXlusjdKJJYJxeIA6xkIP8z9ps2rG7Q7JltKcTyNnkaZLvz6J JIOQ== X-Gm-Message-State: AOAM532HtMNjNfGOauv3W65i5EneSQ5Lf2Ljmi32kAHsMcXxJHYW0UBH cLcNyMeObvKAmZV0R+6gqFr/Wz3JUdHroLnFXZ63KnuWSKiz1m70IkvEsIB9M2t7rp8a6ZfK/08 Tyx8Et4620sNmhgo= X-Received: by 2002:a05:6000:1a85:b0:205:a234:d0a5 with SMTP id f5-20020a0560001a8500b00205a234d0a5mr7325654wry.126.1648805732661; Fri, 01 Apr 2022 02:35:32 -0700 (PDT) X-Google-Smtp-Source: ABdhPJw9RLTvC/AE8rY7DNLnp1JMn6jk32+nRRpi1/LUFKKyc/niR3E5VSQV+Q+PtKlaVjZ7z0bm2Q== X-Received: by 2002:a05:6000:1a85:b0:205:a234:d0a5 with SMTP id f5-20020a0560001a8500b00205a234d0a5mr7325631wry.126.1648805732361; Fri, 01 Apr 2022 02:35:32 -0700 (PDT) Received: from vian.redhat.com ([2a0c:5a80:1c1a:5300:ea85:f79:5235:c7fc]) by smtp.gmail.com with ESMTPSA id az19-20020a05600c601300b0038cadf3aa69sm14009187wmb.36.2022.04.01.02.35.31 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Fri, 01 Apr 2022 02:35:31 -0700 (PDT) From: Nicolas Saenz Julienne To: kwolf@redhat.com, stefanha@redhat.com, berrange@redhat.com Subject: [PATCH v4 3/3] util/event-loop-base: Introduce options to set the thread pool size Date: Fri, 1 Apr 2022 11:35:23 +0200 Message-Id: <20220401093523.873508-4-nsaenzju@redhat.com> X-Mailer: git-send-email 2.35.1 In-Reply-To: <20220401093523.873508-1-nsaenzju@redhat.com> References: <20220401093523.873508-1-nsaenzju@redhat.com> MIME-Version: 1.0 Authentication-Results: relay.mimecast.com; auth=pass smtp.auth=CUSA124A263 smtp.mailfrom=nsaenzju@redhat.com X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset="US-ASCII"; x-default=true Received-SPF: pass client-ip=170.10.129.124; envelope-from=nsaenzju@redhat.com; helo=us-smtp-delivery-124.mimecast.com X-Spam_score_int: -28 X-Spam_score: -2.9 X-Spam_bar: -- X-Spam_report: (-2.9 / 5.0 requ) BAYES_00=-1.9, DKIMWL_WL_HIGH=-0.082, DKIM_SIGNED=0.1, DKIM_VALID=-0.1, DKIM_VALID_AU=-0.1, DKIM_VALID_EF=-0.1, RCVD_IN_DNSWL_LOW=-0.7, RCVD_IN_MSPIKE_H4=0.001, RCVD_IN_MSPIKE_WL=0.001, SPF_HELO_NONE=0.001, SPF_PASS=-0.001, T_SCC_BODY_TEXT_LINE=-0.01 autolearn=ham autolearn_force=no X-Spam_action: no action X-BeenThere: qemu-devel@nongnu.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: fam@euphon.net, eduardo@habkost.net, qemu-block@nongnu.org, michael.roth@amd.com, mtosatti@redhat.com, qemu-devel@nongnu.org, armbru@redhat.com, hreitz@redhat.com, pbonzini@redhat.com, Nicolas Saenz Julienne , eblake@redhat.com Errors-To: qemu-devel-bounces+qemu-devel=archiver.kernel.org@nongnu.org Sender: "Qemu-devel" The thread pool regulates itself: when idle, it kills threads until empty, when in demand, it creates new threads until full. This behaviour doesn't play well with latency sensitive workloads where the price of creating a new thread is too high. For example, when paired with qemu's '-mlock', or using safety features like SafeStack, creating a new thread has been measured take multiple milliseconds. In order to mitigate this let's introduce a new 'EventLoopBase' property to set the thread pool size. The threads will be created during the pool's initialization or upon updating the property's value, remain available during its lifetime regardless of demand, and destroyed upon freeing it. A properly characterized workload will then be able to configure the pool to avoid any latency spikes. Signed-off-by: Nicolas Saenz Julienne Reviewed-by: Stefan Hajnoczi --- Changes since v3: - Rework qom.json to avoid duplication Changes since v2: - Don't wait when decreasing pool size - Fix qapi versioning Changes since v1: - Add INT_MAX check - Have copy of thread pool sizes in AioContext to properly decouple both instances - More coherent variable naming - Handle case where max_threads decreases - Code comments event-loop-base.c | 23 +++++++++++++ include/block/aio.h | 10 ++++++ include/block/thread-pool.h | 3 ++ include/sysemu/event-loop-base.h | 4 +++ iothread.c | 3 ++ qapi/qom.json | 10 +++++- util/aio-posix.c | 1 + util/async.c | 20 ++++++++++++ util/main-loop.c | 9 ++++++ util/thread-pool.c | 55 +++++++++++++++++++++++++++++--- 10 files changed, 133 insertions(+), 5 deletions(-) diff --git a/event-loop-base.c b/event-loop-base.c index e7f99a6ec8..d5be4dc6fc 100644 --- a/event-loop-base.c +++ b/event-loop-base.c @@ -14,6 +14,7 @@ #include "qemu/osdep.h" #include "qom/object_interfaces.h" #include "qapi/error.h" +#include "block/thread-pool.h" #include "sysemu/event-loop-base.h" typedef struct { @@ -21,9 +22,22 @@ typedef struct { ptrdiff_t offset; /* field's byte offset in EventLoopBase struct */ } EventLoopBaseParamInfo; +static void event_loop_base_instance_init(Object *obj) +{ + EventLoopBase *base = EVENT_LOOP_BASE(obj); + + base->thread_pool_max = THREAD_POOL_MAX_THREADS_DEFAULT; +} + static EventLoopBaseParamInfo aio_max_batch_info = { "aio-max-batch", offsetof(EventLoopBase, aio_max_batch), }; +static EventLoopBaseParamInfo thread_pool_min_info = { + "thread-pool-min", offsetof(EventLoopBase, thread_pool_min), +}; +static EventLoopBaseParamInfo thread_pool_max_info = { + "thread-pool-max", offsetof(EventLoopBase, thread_pool_max), +}; static void event_loop_base_get_param(Object *obj, Visitor *v, const char *name, void *opaque, Error **errp) @@ -95,12 +109,21 @@ static void event_loop_base_class_init(ObjectClass *klass, void *class_data) event_loop_base_get_param, event_loop_base_set_param, NULL, &aio_max_batch_info); + object_class_property_add(klass, "thread-pool-min", "int", + event_loop_base_get_param, + event_loop_base_set_param, + NULL, &thread_pool_min_info); + object_class_property_add(klass, "thread-pool-max", "int", + event_loop_base_get_param, + event_loop_base_set_param, + NULL, &thread_pool_max_info); } static const TypeInfo event_loop_base_info = { .name = TYPE_EVENT_LOOP_BASE, .parent = TYPE_OBJECT, .instance_size = sizeof(EventLoopBase), + .instance_init = event_loop_base_instance_init, .class_size = sizeof(EventLoopBaseClass), .class_init = event_loop_base_class_init, .abstract = true, diff --git a/include/block/aio.h b/include/block/aio.h index 5634173b12..d128558f1d 100644 --- a/include/block/aio.h +++ b/include/block/aio.h @@ -192,6 +192,8 @@ struct AioContext { QSLIST_HEAD(, Coroutine) scheduled_coroutines; QEMUBH *co_schedule_bh; + int thread_pool_min; + int thread_pool_max; /* Thread pool for performing work and receiving completion callbacks. * Has its own locking. */ @@ -769,4 +771,12 @@ void aio_context_set_poll_params(AioContext *ctx, int64_t max_ns, void aio_context_set_aio_params(AioContext *ctx, int64_t max_batch, Error **errp); +/** + * aio_context_set_thread_pool_params: + * @ctx: the aio context + * @min: min number of threads to have readily available in the thread pool + * @min: max number of threads the thread pool can contain + */ +void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min, + int64_t max, Error **errp); #endif diff --git a/include/block/thread-pool.h b/include/block/thread-pool.h index 7dd7d730a0..2020bcc92d 100644 --- a/include/block/thread-pool.h +++ b/include/block/thread-pool.h @@ -20,6 +20,8 @@ #include "block/block.h" +#define THREAD_POOL_MAX_THREADS_DEFAULT 64 + typedef int ThreadPoolFunc(void *opaque); typedef struct ThreadPool ThreadPool; @@ -33,5 +35,6 @@ BlockAIOCB *thread_pool_submit_aio(ThreadPool *pool, int coroutine_fn thread_pool_submit_co(ThreadPool *pool, ThreadPoolFunc *func, void *arg); void thread_pool_submit(ThreadPool *pool, ThreadPoolFunc *func, void *arg); +void thread_pool_update_params(ThreadPool *pool, struct AioContext *ctx); #endif diff --git a/include/sysemu/event-loop-base.h b/include/sysemu/event-loop-base.h index fced4c9fea..2748bf6ae1 100644 --- a/include/sysemu/event-loop-base.h +++ b/include/sysemu/event-loop-base.h @@ -33,5 +33,9 @@ struct EventLoopBase { /* AioContext AIO engine parameters */ int64_t aio_max_batch; + + /* AioContext thread pool parameters */ + int64_t thread_pool_min; + int64_t thread_pool_max; }; #endif diff --git a/iothread.c b/iothread.c index 8fa2f3bfb8..529194a566 100644 --- a/iothread.c +++ b/iothread.c @@ -174,6 +174,9 @@ static void iothread_set_aio_context_params(EventLoopBase *base, Error **errp) aio_context_set_aio_params(iothread->ctx, iothread->parent_obj.aio_max_batch, errp); + + aio_context_set_thread_pool_params(iothread->ctx, base->thread_pool_min, + base->thread_pool_max, errp); } diff --git a/qapi/qom.json b/qapi/qom.json index e5f31c4469..06b8c3d10b 100644 --- a/qapi/qom.json +++ b/qapi/qom.json @@ -506,9 +506,17 @@ # # @aio-max-batch: maximum number of requests in a batch for the AIO engine, # 0 means that the engine will use its default. +# +# @thread-pool-min: minimum number of threads readily available in the thread +# pool (default:0, since 7.1) +# +# @thread-pool-max: maximum number of threads the thread pool can contain +# (default:64, since 7.1) ## { 'struct': 'EventLoopBaseProperties', - 'data': { '*aio-max-batch': 'int' } } + 'data': { '*aio-max-batch': 'int', + '*thread-pool-min': 'int', + '*thread-pool-max': 'int' } } ## # @IothreadProperties: diff --git a/util/aio-posix.c b/util/aio-posix.c index be0182a3c6..731f3826c0 100644 --- a/util/aio-posix.c +++ b/util/aio-posix.c @@ -15,6 +15,7 @@ #include "qemu/osdep.h" #include "block/block.h" +#include "block/thread-pool.h" #include "qemu/main-loop.h" #include "qemu/rcu.h" #include "qemu/rcu_queue.h" diff --git a/util/async.c b/util/async.c index 2ea1172f3e..554ba70cca 100644 --- a/util/async.c +++ b/util/async.c @@ -563,6 +563,9 @@ AioContext *aio_context_new(Error **errp) ctx->aio_max_batch = 0; + ctx->thread_pool_min = 0; + ctx->thread_pool_max = THREAD_POOL_MAX_THREADS_DEFAULT; + return ctx; fail: g_source_destroy(&ctx->source); @@ -696,3 +699,20 @@ void qemu_set_current_aio_context(AioContext *ctx) assert(!get_my_aiocontext()); set_my_aiocontext(ctx); } + +void aio_context_set_thread_pool_params(AioContext *ctx, int64_t min, + int64_t max, Error **errp) +{ + + if (min > max || !max || min > INT_MAX || max > INT_MAX) { + error_setg(errp, "bad thread-pool-min/thread-pool-max values"); + return; + } + + ctx->thread_pool_min = min; + ctx->thread_pool_max = max; + + if (ctx->thread_pool) { + thread_pool_update_params(ctx->thread_pool, ctx); + } +} diff --git a/util/main-loop.c b/util/main-loop.c index 5b13f456fa..a0f48186ab 100644 --- a/util/main-loop.c +++ b/util/main-loop.c @@ -30,6 +30,7 @@ #include "sysemu/replay.h" #include "qemu/main-loop.h" #include "block/aio.h" +#include "block/thread-pool.h" #include "qemu/error-report.h" #include "qemu/queue.h" #include "qemu/compiler.h" @@ -187,12 +188,20 @@ int qemu_init_main_loop(Error **errp) static void main_loop_update_params(EventLoopBase *base, Error **errp) { + ERRP_GUARD(); + if (!qemu_aio_context) { error_setg(errp, "qemu aio context not ready"); return; } aio_context_set_aio_params(qemu_aio_context, base->aio_max_batch, errp); + if (*errp) { + return; + } + + aio_context_set_thread_pool_params(qemu_aio_context, base->thread_pool_min, + base->thread_pool_max, errp); } MainLoop *mloop; diff --git a/util/thread-pool.c b/util/thread-pool.c index d763cea505..196835b4d3 100644 --- a/util/thread-pool.c +++ b/util/thread-pool.c @@ -58,7 +58,6 @@ struct ThreadPool { QemuMutex lock; QemuCond worker_stopped; QemuSemaphore sem; - int max_threads; QEMUBH *new_thread_bh; /* The following variables are only accessed from one AioContext. */ @@ -71,8 +70,27 @@ struct ThreadPool { int new_threads; /* backlog of threads we need to create */ int pending_threads; /* threads created but not running yet */ bool stopping; + int min_threads; + int max_threads; }; +static inline bool back_to_sleep(ThreadPool *pool, int ret) +{ + /* + * The semaphore timed out, we should exit the loop except when: + * - There is work to do, we raced with the signal. + * - The max threads threshold just changed, we raced with the signal. + * - The thread pool forces a minimum number of readily available threads. + */ + if (ret == -1 && (!QTAILQ_EMPTY(&pool->request_list) || + pool->cur_threads > pool->max_threads || + pool->cur_threads <= pool->min_threads)) { + return true; + } + + return false; +} + static void *worker_thread(void *opaque) { ThreadPool *pool = opaque; @@ -91,8 +109,9 @@ static void *worker_thread(void *opaque) ret = qemu_sem_timedwait(&pool->sem, 10000); qemu_mutex_lock(&pool->lock); pool->idle_threads--; - } while (ret == -1 && !QTAILQ_EMPTY(&pool->request_list)); - if (ret == -1 || pool->stopping) { + } while (back_to_sleep(pool, ret)); + if (ret == -1 || pool->stopping || + pool->cur_threads > pool->max_threads) { break; } @@ -294,6 +313,33 @@ void thread_pool_submit(ThreadPool *pool, ThreadPoolFunc *func, void *arg) thread_pool_submit_aio(pool, func, arg, NULL, NULL); } +void thread_pool_update_params(ThreadPool *pool, AioContext *ctx) +{ + qemu_mutex_lock(&pool->lock); + + pool->min_threads = ctx->thread_pool_min; + pool->max_threads = ctx->thread_pool_max; + + /* + * We either have to: + * - Increase the number available of threads until over the min_threads + * threshold. + * - Decrease the number of available threads until under the max_threads + * threshold. + * - Do nothing. The current number of threads fall in between the min and + * max thresholds. We'll let the pool manage itself. + */ + for (int i = pool->cur_threads; i < pool->min_threads; i++) { + spawn_thread(pool); + } + + for (int i = pool->cur_threads; i > pool->max_threads; i--) { + qemu_sem_post(&pool->sem); + } + + qemu_mutex_unlock(&pool->lock); +} + static void thread_pool_init_one(ThreadPool *pool, AioContext *ctx) { if (!ctx) { @@ -306,11 +352,12 @@ static void thread_pool_init_one(ThreadPool *pool, AioContext *ctx) qemu_mutex_init(&pool->lock); qemu_cond_init(&pool->worker_stopped); qemu_sem_init(&pool->sem, 0); - pool->max_threads = 64; pool->new_thread_bh = aio_bh_new(ctx, spawn_thread_bh_fn, pool); QLIST_INIT(&pool->head); QTAILQ_INIT(&pool->request_list); + + thread_pool_update_params(pool, ctx); } ThreadPool *thread_pool_new(AioContext *ctx) -- 2.35.1