From patchwork Mon Sep 6 14:59:46 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Anton Ivanov X-Patchwork-Id: 1524984 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (sender SPF authorized) smtp.mailfrom=openvswitch.org (client-ip=140.211.166.138; helo=smtp1.osuosl.org; envelope-from=ovs-dev-bounces@openvswitch.org; receiver=) Received: from smtp1.osuosl.org (smtp1.osuosl.org [140.211.166.138]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 4H3BQV4SsDz9sW5 for ; Tue, 7 Sep 2021 01:00:02 +1000 (AEST) Received: from localhost (localhost [127.0.0.1]) by smtp1.osuosl.org (Postfix) with ESMTP id CFA3480E51; Mon, 6 Sep 2021 15:00:00 +0000 (UTC) X-Virus-Scanned: amavisd-new at osuosl.org Received: from smtp1.osuosl.org ([127.0.0.1]) by localhost (smtp1.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 91DfO39BqJPo; Mon, 6 Sep 2021 14:59:59 +0000 (UTC) Received: from lists.linuxfoundation.org (lf-lists.osuosl.org [140.211.9.56]) by smtp1.osuosl.org (Postfix) with ESMTPS id A174480BEE; Mon, 6 Sep 2021 14:59:58 +0000 (UTC) Received: from lf-lists.osuosl.org (localhost [127.0.0.1]) by lists.linuxfoundation.org (Postfix) with ESMTP id 72E60C001D; Mon, 6 Sep 2021 14:59:58 +0000 (UTC) X-Original-To: ovs-dev@openvswitch.org Delivered-To: ovs-dev@lists.linuxfoundation.org Received: from smtp1.osuosl.org (smtp1.osuosl.org [140.211.166.138]) by lists.linuxfoundation.org (Postfix) with ESMTP id 09CBCC001B for ; Mon, 6 Sep 2021 14:59:58 +0000 (UTC) Received: from localhost (localhost [127.0.0.1]) by smtp1.osuosl.org (Postfix) with ESMTP id DA10080C20 for ; Mon, 6 Sep 2021 14:59:57 +0000 (UTC) X-Virus-Scanned: amavisd-new at osuosl.org Received: from smtp1.osuosl.org ([127.0.0.1]) by localhost (smtp1.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id NB58ipO4ufzE for ; Mon, 6 Sep 2021 14:59:56 +0000 (UTC) X-Greylist: from auto-whitelisted by SQLgrey-1.8.0 Received: from www.kot-begemot.co.uk (ivanoab7.miniserver.com [37.128.132.42]) by smtp1.osuosl.org (Postfix) with ESMTPS id D5FB680BFC for ; Mon, 6 Sep 2021 14:59:55 +0000 (UTC) Received: from tun252.jain.kot-begemot.co.uk ([192.168.18.6] helo=jain.kot-begemot.co.uk) by www.kot-begemot.co.uk with esmtps (TLS1.3:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.92) (envelope-from ) id 1mNG6C-0001Pi-BF for ovs-dev@openvswitch.org; Mon, 06 Sep 2021 14:59:52 +0000 Received: from jain.kot-begemot.co.uk ([192.168.3.3]) by jain.kot-begemot.co.uk with esmtp (Exim 4.92) (envelope-from ) id 1mNG68-0004ph-N8; Mon, 06 Sep 2021 15:59:50 +0100 From: anton.ivanov@cambridgegreys.com To: ovs-dev@openvswitch.org Date: Mon, 6 Sep 2021 15:59:46 +0100 Message-Id: <20210906145947.18521-1-anton.ivanov@cambridgegreys.com> X-Mailer: git-send-email 2.20.1 MIME-Version: 1.0 X-Clacks-Overhead: GNU Terry Pratchett Cc: Anton Ivanov Subject: [ovs-dev] [OVN Patch v3 1/2] Make changes to the parallel processing API to allow pool sizing X-BeenThere: ovs-dev@openvswitch.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: ovs-dev-bounces@openvswitch.org Sender: "dev" From: Anton Ivanov 1. Make pool size user defineable. 2. Expose pool destruction. 3. Make pools resizeable at runtime. Signed-off-by: Anton Ivanov --- lib/ovn-parallel-hmap.c | 202 ++++++++++++++++++++++++++++++---------- lib/ovn-parallel-hmap.h | 23 ++++- northd/ovn-northd.c | 58 +++++------- ovs | 2 +- 4 files changed, 194 insertions(+), 91 deletions(-) diff --git a/lib/ovn-parallel-hmap.c b/lib/ovn-parallel-hmap.c index b8c7ac786..30de457b5 100644 --- a/lib/ovn-parallel-hmap.c +++ b/lib/ovn-parallel-hmap.c @@ -51,7 +51,6 @@ static bool can_parallelize = false; * accompanied by a fence. It does not need to be atomic or be * accessed under a lock. */ -static bool workers_must_exit = false; static struct ovs_list worker_pools = OVS_LIST_INITIALIZER(&worker_pools); @@ -70,10 +69,20 @@ static void merge_hash_results(struct worker_pool *pool OVS_UNUSED, void *fin_result, void *result_frags, int index); + +static bool init_control(struct worker_control *control, int id, + struct worker_pool *pool); + +static void cleanup_control(struct worker_pool *pool, int id); + +static void free_controls(struct worker_pool *pool); + +static struct worker_control *alloc_controls(int size); + bool -ovn_stop_parallel_processing(void) +ovn_stop_parallel_processing(struct worker_pool *pool) { - return workers_must_exit; + return pool->workers_must_exit; } bool @@ -92,11 +101,67 @@ ovn_can_parallelize_hashes(bool force_parallel) return can_parallelize; } + +void +destroy_pool(struct worker_pool *pool) { + char sem_name[256]; + + free_controls(pool); + sem_close(pool->done); + sprintf(sem_name, MAIN_SEM_NAME, sembase, pool); + sem_unlink(sem_name); + free(pool); +} + +bool +ovn_resize_pool(struct worker_pool *pool, int size) +{ + int i; + + ovs_assert(pool != NULL); + + if (!size) { + size = pool_size; + } + + ovs_mutex_lock(&init_mutex); + + if (can_parallelize) { + free_controls(pool); + pool->size = size; + + /* Allocate new control structures. */ + + pool->controls = alloc_controls(size); + pool->workers_must_exit = false; + + for (i = 0; i < pool->size; i++) { + if (! init_control(&pool->controls[i], i, pool)) { + goto cleanup; + } + } + } + ovs_mutex_unlock(&init_mutex); + return true; +cleanup: + + /* Something went wrong when opening semaphores. In this case + * it is better to shut off parallel procesing altogether + */ + + VLOG_INFO("Failed to initialize parallel processing, error %d", errno); + can_parallelize = false; + free_controls(pool); + + ovs_mutex_unlock(&init_mutex); + return false; +} + + struct worker_pool * -ovn_add_worker_pool(void *(*start)(void *)) +ovn_add_worker_pool(void *(*start)(void *), int size) { struct worker_pool *new_pool = NULL; - struct worker_control *new_control; bool test = false; int i; char sem_name[256]; @@ -113,38 +178,29 @@ ovn_add_worker_pool(void *(*start)(void *)) ovs_mutex_unlock(&init_mutex); } + if (!size) { + size = pool_size; + } + ovs_mutex_lock(&init_mutex); if (can_parallelize) { new_pool = xmalloc(sizeof(struct worker_pool)); - new_pool->size = pool_size; - new_pool->controls = NULL; + new_pool->size = size; + new_pool->start = start; sprintf(sem_name, MAIN_SEM_NAME, sembase, new_pool); new_pool->done = sem_open(sem_name, O_CREAT, S_IRWXU, 0); if (new_pool->done == SEM_FAILED) { goto cleanup; } - new_pool->controls = - xmalloc(sizeof(struct worker_control) * new_pool->size); + new_pool->controls = alloc_controls(size); + new_pool->workers_must_exit = false; for (i = 0; i < new_pool->size; i++) { - new_control = &new_pool->controls[i]; - new_control->id = i; - new_control->done = new_pool->done; - new_control->data = NULL; - ovs_mutex_init(&new_control->mutex); - new_control->finished = ATOMIC_VAR_INIT(false); - sprintf(sem_name, WORKER_SEM_NAME, sembase, new_pool, i); - new_control->fire = sem_open(sem_name, O_CREAT, S_IRWXU, 0); - if (new_control->fire == SEM_FAILED) { + if (!init_control(&new_pool->controls[i], i, new_pool)) { goto cleanup; } } - - for (i = 0; i < pool_size; i++) { - new_pool->controls[i].worker = - ovs_thread_create("worker pool helper", start, &new_pool->controls[i]); - } ovs_list_push_back(&worker_pools, &new_pool->list_node); } ovs_mutex_unlock(&init_mutex); @@ -157,16 +213,7 @@ cleanup: VLOG_INFO("Failed to initialize parallel processing, error %d", errno); can_parallelize = false; - if (new_pool->controls) { - for (i = 0; i < new_pool->size; i++) { - if (new_pool->controls[i].fire != SEM_FAILED) { - sem_close(new_pool->controls[i].fire); - sprintf(sem_name, WORKER_SEM_NAME, sembase, new_pool, i); - sem_unlink(sem_name); - break; /* semaphores past this one are uninitialized */ - } - } - } + free_controls(new_pool); if (new_pool->done != SEM_FAILED) { sem_close(new_pool->done); sprintf(sem_name, MAIN_SEM_NAME, sembase, new_pool); @@ -176,7 +223,6 @@ cleanup: return NULL; } - /* Initializes 'hmap' as an empty hash table with mask N. */ void ovn_fast_hmap_init(struct hmap *hmap, ssize_t mask) @@ -365,14 +411,84 @@ ovn_update_hashrow_locks(struct hmap *lflows, struct hashrow_locks *hrl) } } +static bool +init_control(struct worker_control *control, int id, + struct worker_pool *pool) +{ + char sem_name[256]; + control->id = id; + control->done = pool->done; + control->data = NULL; + ovs_mutex_init(&control->mutex); + control->finished = ATOMIC_VAR_INIT(false); + sprintf(sem_name, WORKER_SEM_NAME, sembase, pool, id); + control->fire = sem_open(sem_name, O_CREAT, S_IRWXU, 0); + control->pool = pool; + control->worker = 0; + if (control->fire == SEM_FAILED) { + return false; + } + control->worker = + ovs_thread_create("worker pool helper", pool->start, control); + return true; +} + static void -worker_pool_hook(void *aux OVS_UNUSED) { +cleanup_control(struct worker_pool *pool, int id) +{ + char sem_name[256]; + struct worker_control *control = &pool->controls[id]; + + if (control->fire != SEM_FAILED) { + sem_close(control->fire); + sprintf(sem_name, WORKER_SEM_NAME, sembase, pool, id); + sem_unlink(sem_name); + } +} + +static void +free_controls(struct worker_pool *pool) +{ int i; + if (pool->controls) { + pool->workers_must_exit = true; + for (i = 0; i < pool->size ; i++) { + if (pool->controls[i].fire != SEM_FAILED) { + sem_post(pool->controls[i].fire); + } + } + for (i = 0; i < pool->size ; i++) { + if (pool->controls[i].worker) { + pthread_join(pool->controls[i].worker, NULL); + pool->controls[i].worker = 0; + } + } + for (i = 0; i < pool->size; i++) { + cleanup_control(pool, i); + } + free(pool->controls); + pool->controls = NULL; + pool->workers_must_exit = false; + } +} + +static struct worker_control *alloc_controls(int size) +{ + int i; + struct worker_control *controls = + xcalloc(sizeof(struct worker_control), size); + + for (i = 0; i < size ; i++) { + controls[i].fire = SEM_FAILED; + } + return controls; +} + +static void +worker_pool_hook(void *aux OVS_UNUSED) { static struct worker_pool *pool; char sem_name[256]; - workers_must_exit = true; - /* All workers must honour the must_exit flag and check for it regularly. * We can make it atomic and check it via atomics in workers, but that * is not really necessary as it is set just once - when the program @@ -383,17 +499,7 @@ worker_pool_hook(void *aux OVS_UNUSED) { /* Wake up the workers after the must_exit flag has been set */ LIST_FOR_EACH (pool, list_node, &worker_pools) { - for (i = 0; i < pool->size ; i++) { - sem_post(pool->controls[i].fire); - } - for (i = 0; i < pool->size ; i++) { - pthread_join(pool->controls[i].worker, NULL); - } - for (i = 0; i < pool->size ; i++) { - sem_close(pool->controls[i].fire); - sprintf(sem_name, WORKER_SEM_NAME, sembase, pool, i); - sem_unlink(sem_name); - } + free_controls(pool); sem_close(pool->done); sprintf(sem_name, MAIN_SEM_NAME, sembase, pool); sem_unlink(sem_name); diff --git a/lib/ovn-parallel-hmap.h b/lib/ovn-parallel-hmap.h index 2df132ea8..4708f41f2 100644 --- a/lib/ovn-parallel-hmap.h +++ b/lib/ovn-parallel-hmap.h @@ -83,6 +83,7 @@ struct worker_control { void *data; /* Pointer to data to be processed. */ void *workload; /* back-pointer to the worker pool structure. */ pthread_t worker; + struct worker_pool *pool; }; struct worker_pool { @@ -90,16 +91,21 @@ struct worker_pool { struct ovs_list list_node; /* List of pools - used in cleanup/exit. */ struct worker_control *controls; /* "Handles" in this pool. */ sem_t *done; /* Work completion semaphorew. */ + void *(*start)(void *); /* Work function. */ + bool workers_must_exit; /* Pool to be destroyed flag. */ }; /* Add a worker pool for thread function start() which expects a pointer to - * a worker_control structure as an argument. */ + * a worker_control structure as an argument. + * If size is non-zero, it is used for pool sizing. If size is zero, pool + * size uses system defaults. + */ -struct worker_pool *ovn_add_worker_pool(void *(*start)(void *)); +struct worker_pool *ovn_add_worker_pool(void *(*start)(void *), int size); /* Setting this to true will make all processing threads exit */ -bool ovn_stop_parallel_processing(void); +bool ovn_stop_parallel_processing(struct worker_pool *pool); /* Build a hmap pre-sized for size elements */ @@ -253,6 +259,10 @@ static inline void init_hash_row_locks(struct hashrow_locks *hrl) bool ovn_can_parallelize_hashes(bool force_parallel); +void ovn_destroy_pool(struct worker_pool *pool); + +bool ovn_resize_pool(struct worker_pool *pool, int size); + /* Use the OVN library functions for stuff which OVS has not defined * If OVS has defined these, they will still compile using the OVN * local names, but will be dropped by the linker in favour of the OVS @@ -263,9 +273,9 @@ bool ovn_can_parallelize_hashes(bool force_parallel); #define can_parallelize_hashes(force) ovn_can_parallelize_hashes(force) -#define stop_parallel_processing() ovn_stop_parallel_processing() +#define stop_parallel_processing(pool) ovn_stop_parallel_processing(pool) -#define add_worker_pool(start) ovn_add_worker_pool(start) +#define add_worker_pool(start, size) ovn_add_worker_pool(start, size) #define fast_hmap_size_for(hmap, size) ovn_fast_hmap_size_for(hmap, size) @@ -286,6 +296,9 @@ bool ovn_can_parallelize_hashes(bool force_parallel); #define run_pool_callback(pool, fin_result, result_frags, helper_func) \ ovn_run_pool_callback(pool, fin_result, result_frags, helper_func) +#define destroy_pool(pool) ovn_destroy_pool(pool) + +#define resize_pool(pool, size) ovn_resize_pool(pool, size) #ifdef __clang__ diff --git a/northd/ovn-northd.c b/northd/ovn-northd.c index ee761cef0..324800c32 100644 --- a/northd/ovn-northd.c +++ b/northd/ovn-northd.c @@ -12828,16 +12828,10 @@ build_lswitch_and_lrouter_iterate_by_op(struct ovn_port *op, &lsi->actions); } -struct lflows_thread_pool { - struct worker_pool *pool; -}; - - static void * build_lflows_thread(void *arg) { struct worker_control *control = (struct worker_control *) arg; - struct lflows_thread_pool *workload; struct lswitch_flow_build_info *lsi; struct ovn_datapath *od; @@ -12846,21 +12840,21 @@ build_lflows_thread(void *arg) struct ovn_igmp_group *igmp_group; int bnum; - while (!stop_parallel_processing()) { + + while (!stop_parallel_processing(control->pool)) { wait_for_work(control); - workload = (struct lflows_thread_pool *) control->workload; lsi = (struct lswitch_flow_build_info *) control->data; - if (stop_parallel_processing()) { + if (stop_parallel_processing(control->pool)) { return NULL; } - if (lsi && workload) { + if (lsi) { /* Iterate over bucket ThreadID, ThreadID+size, ... */ for (bnum = control->id; bnum <= lsi->datapaths->mask; - bnum += workload->pool->size) + bnum += control->pool->size) { HMAP_FOR_EACH_IN_PARALLEL (od, key_node, bnum, lsi->datapaths) { - if (stop_parallel_processing()) { + if (stop_parallel_processing(control->pool)) { return NULL; } build_lswitch_and_lrouter_iterate_by_od(od, lsi); @@ -12868,10 +12862,10 @@ build_lflows_thread(void *arg) } for (bnum = control->id; bnum <= lsi->ports->mask; - bnum += workload->pool->size) + bnum += control->pool->size) { HMAP_FOR_EACH_IN_PARALLEL (op, key_node, bnum, lsi->ports) { - if (stop_parallel_processing()) { + if (stop_parallel_processing(control->pool)) { return NULL; } build_lswitch_and_lrouter_iterate_by_op(op, lsi); @@ -12879,10 +12873,10 @@ build_lflows_thread(void *arg) } for (bnum = control->id; bnum <= lsi->lbs->mask; - bnum += workload->pool->size) + bnum += control->pool->size) { HMAP_FOR_EACH_IN_PARALLEL (lb, hmap_node, bnum, lsi->lbs) { - if (stop_parallel_processing()) { + if (stop_parallel_processing(control->pool)) { return NULL; } build_lswitch_arp_nd_service_monitor(lb, lsi->lflows, @@ -12900,11 +12894,11 @@ build_lflows_thread(void *arg) } for (bnum = control->id; bnum <= lsi->igmp_groups->mask; - bnum += workload->pool->size) + bnum += control->pool->size) { HMAP_FOR_EACH_IN_PARALLEL ( igmp_group, hmap_node, bnum, lsi->igmp_groups) { - if (stop_parallel_processing()) { + if (stop_parallel_processing(control->pool)) { return NULL; } build_lswitch_ip_mcast_igmp_mld(igmp_group, lsi->lflows, @@ -12919,24 +12913,14 @@ build_lflows_thread(void *arg) } static bool pool_init_done = false; -static struct lflows_thread_pool *build_lflows_pool = NULL; +static struct worker_pool *build_lflows_pool = NULL; static void init_lflows_thread_pool(void) { - int index; - if (!pool_init_done) { - struct worker_pool *pool = add_worker_pool(build_lflows_thread); + build_lflows_pool = add_worker_pool(build_lflows_thread, 0); pool_init_done = true; - if (pool) { - build_lflows_pool = xmalloc(sizeof(*build_lflows_pool)); - build_lflows_pool->pool = pool; - for (index = 0; index < build_lflows_pool->pool->size; index++) { - build_lflows_pool->pool->controls[index].workload = - build_lflows_pool; - } - } } } @@ -12979,16 +12963,16 @@ build_lswitch_and_lrouter_flows(struct hmap *datapaths, struct hmap *ports, struct lswitch_flow_build_info *lsiv; int index; - lsiv = xcalloc(sizeof(*lsiv), build_lflows_pool->pool->size); + lsiv = xcalloc(sizeof(*lsiv), build_lflows_pool->size); if (use_logical_dp_groups) { lflow_segs = NULL; } else { - lflow_segs = xcalloc(sizeof(*lflow_segs), build_lflows_pool->pool->size); + lflow_segs = xcalloc(sizeof(*lflow_segs), build_lflows_pool->size); } /* Set up "work chunks" for each thread to work on. */ - for (index = 0; index < build_lflows_pool->pool->size; index++) { + for (index = 0; index < build_lflows_pool->size; index++) { if (use_logical_dp_groups) { /* if dp_groups are in use we lock a shared lflows hash * on a per-bucket level instead of merging hash frags */ @@ -13010,17 +12994,17 @@ build_lswitch_and_lrouter_flows(struct hmap *datapaths, struct hmap *ports, ds_init(&lsiv[index].match); ds_init(&lsiv[index].actions); - build_lflows_pool->pool->controls[index].data = &lsiv[index]; + build_lflows_pool->controls[index].data = &lsiv[index]; } /* Run thread pool. */ if (use_logical_dp_groups) { - run_pool_callback(build_lflows_pool->pool, NULL, NULL, noop_callback); + run_pool_callback(build_lflows_pool, NULL, NULL, noop_callback); } else { - run_pool_hash(build_lflows_pool->pool, lflows, lflow_segs); + run_pool_hash(build_lflows_pool, lflows, lflow_segs); } - for (index = 0; index < build_lflows_pool->pool->size; index++) { + for (index = 0; index < build_lflows_pool->size; index++) { ds_destroy(&lsiv[index].match); ds_destroy(&lsiv[index].actions); } diff --git a/ovs b/ovs index 748010ff3..50e5523b9 160000 --- a/ovs +++ b/ovs @@ -1 +1 @@ -Subproject commit 748010ff304b7cd2c43f4eb98a554433f0df07f9 +Subproject commit 50e5523b9b2b154e5fafc5acdcdec85e9cc5a330 From patchwork Mon Sep 6 14:59:47 2021 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Anton Ivanov X-Patchwork-Id: 1524985 Return-Path: X-Original-To: incoming@patchwork.ozlabs.org Delivered-To: patchwork-incoming@bilbo.ozlabs.org Authentication-Results: ozlabs.org; spf=pass (sender SPF authorized) smtp.mailfrom=openvswitch.org (client-ip=2605:bc80:3010::138; helo=smtp1.osuosl.org; envelope-from=ovs-dev-bounces@openvswitch.org; receiver=) Received: from smtp1.osuosl.org (smtp1.osuosl.org [IPv6:2605:bc80:3010::138]) (using TLSv1.3 with cipher TLS_AES_256_GCM_SHA384 (256/256 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256) (No client certificate requested) by ozlabs.org (Postfix) with ESMTPS id 4H3BQb4Q8Sz9sW5 for ; Tue, 7 Sep 2021 01:00:07 +1000 (AEST) Received: from localhost (localhost [127.0.0.1]) by smtp1.osuosl.org (Postfix) with ESMTP id 2D00181B53; Mon, 6 Sep 2021 15:00:05 +0000 (UTC) X-Virus-Scanned: amavisd-new at osuosl.org Received: from smtp1.osuosl.org ([127.0.0.1]) by localhost (smtp1.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id 1PBS5iSOJ3Hv; Mon, 6 Sep 2021 15:00:03 +0000 (UTC) Received: from lists.linuxfoundation.org (lf-lists.osuosl.org [140.211.9.56]) by smtp1.osuosl.org (Postfix) with ESMTPS id 7E98080F85; Mon, 6 Sep 2021 15:00:02 +0000 (UTC) Received: from lf-lists.osuosl.org (localhost [127.0.0.1]) by lists.linuxfoundation.org (Postfix) with ESMTP id 552AEC001D; Mon, 6 Sep 2021 15:00:02 +0000 (UTC) X-Original-To: ovs-dev@openvswitch.org Delivered-To: ovs-dev@lists.linuxfoundation.org Received: from smtp2.osuosl.org (smtp2.osuosl.org [140.211.166.133]) by lists.linuxfoundation.org (Postfix) with ESMTP id 7B341C0021 for ; Mon, 6 Sep 2021 14:59:58 +0000 (UTC) Received: from localhost (localhost [127.0.0.1]) by smtp2.osuosl.org (Postfix) with ESMTP id 5EF09401C3 for ; Mon, 6 Sep 2021 14:59:58 +0000 (UTC) X-Virus-Scanned: amavisd-new at osuosl.org Received: from smtp2.osuosl.org ([127.0.0.1]) by localhost (smtp2.osuosl.org [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id MtDJH8I9BxyX for ; Mon, 6 Sep 2021 14:59:57 +0000 (UTC) X-Greylist: from auto-whitelisted by SQLgrey-1.8.0 Received: from www.kot-begemot.co.uk (ivanoab7.miniserver.com [37.128.132.42]) by smtp2.osuosl.org (Postfix) with ESMTPS id C7AB140109 for ; Mon, 6 Sep 2021 14:59:56 +0000 (UTC) Received: from tun252.jain.kot-begemot.co.uk ([192.168.18.6] helo=jain.kot-begemot.co.uk) by www.kot-begemot.co.uk with esmtps (TLS1.3:ECDHE_RSA_AES_256_GCM_SHA384:256) (Exim 4.92) (envelope-from ) id 1mNG6E-0001Pn-Dv for ovs-dev@openvswitch.org; Mon, 06 Sep 2021 14:59:54 +0000 Received: from jain.kot-begemot.co.uk ([192.168.3.3]) by jain.kot-begemot.co.uk with esmtp (Exim 4.92) (envelope-from ) id 1mNG6A-0004ph-Pv; Mon, 06 Sep 2021 15:59:52 +0100 From: anton.ivanov@cambridgegreys.com To: ovs-dev@openvswitch.org Date: Mon, 6 Sep 2021 15:59:47 +0100 Message-Id: <20210906145947.18521-2-anton.ivanov@cambridgegreys.com> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20210906145947.18521-1-anton.ivanov@cambridgegreys.com> References: <20210906145947.18521-1-anton.ivanov@cambridgegreys.com> MIME-Version: 1.0 X-Clacks-Overhead: GNU Terry Pratchett Cc: Anton Ivanov Subject: [ovs-dev] [OVN Patch v3 2/2] Add support for configuring parallelization via unixctl X-BeenThere: ovs-dev@openvswitch.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: ovs-dev-bounces@openvswitch.org Sender: "dev" From: Anton Ivanov Signed-off-by: Anton Ivanov --- lib/ovn-parallel-hmap.c | 209 ++++++++++++++++++++++++++++++++++++++-- lib/ovn-parallel-hmap.h | 63 +++++++++++- northd/ovn-northd.c | 26 ++--- tests/ovn-macros.at | 16 ++- 4 files changed, 283 insertions(+), 31 deletions(-) diff --git a/lib/ovn-parallel-hmap.c b/lib/ovn-parallel-hmap.c index 30de457b5..8a055b2c6 100644 --- a/lib/ovn-parallel-hmap.c +++ b/lib/ovn-parallel-hmap.c @@ -33,6 +33,7 @@ #include "ovs-thread.h" #include "ovs-numa.h" #include "random.h" +#include "unixctl.h" VLOG_DEFINE_THIS_MODULE(ovn_parallel_hmap); @@ -46,6 +47,7 @@ VLOG_DEFINE_THIS_MODULE(ovn_parallel_hmap); */ static atomic_bool initial_pool_setup = ATOMIC_VAR_INIT(false); static bool can_parallelize = false; +static bool should_parallelize = false; /* This is set only in the process of exit and the set is * accompanied by a fence. It does not need to be atomic or be @@ -85,6 +87,19 @@ ovn_stop_parallel_processing(struct worker_pool *pool) return pool->workers_must_exit; } +bool +ovn_set_parallel_processing(bool enable) +{ + should_parallelize = enable; + return can_parallelize; +} + +bool +ovn_get_parallel_processing(void) +{ + return can_parallelize && should_parallelize; +} + bool ovn_can_parallelize_hashes(bool force_parallel) { @@ -110,6 +125,7 @@ destroy_pool(struct worker_pool *pool) { sem_close(pool->done); sprintf(sem_name, MAIN_SEM_NAME, sembase, pool); sem_unlink(sem_name); + free(pool->name); free(pool); } @@ -120,6 +136,10 @@ ovn_resize_pool(struct worker_pool *pool, int size) ovs_assert(pool != NULL); + if (!pool->is_mutable) { + return false; + } + if (!size) { size = pool_size; } @@ -159,7 +179,8 @@ cleanup: struct worker_pool * -ovn_add_worker_pool(void *(*start)(void *), int size) +ovn_add_worker_pool(void *(*start)(void *), int size, char *name, + bool is_mutable) { struct worker_pool *new_pool = NULL; bool test = false; @@ -187,6 +208,8 @@ ovn_add_worker_pool(void *(*start)(void *), int size) new_pool = xmalloc(sizeof(struct worker_pool)); new_pool->size = size; new_pool->start = start; + new_pool->is_mutable = is_mutable; + new_pool->name = xstrdup(name); sprintf(sem_name, MAIN_SEM_NAME, sembase, new_pool); new_pool->done = sem_open(sem_name, O_CREAT, S_IRWXU, 0); if (new_pool->done == SEM_FAILED) { @@ -219,6 +242,7 @@ cleanup: sprintf(sem_name, MAIN_SEM_NAME, sembase, new_pool); sem_unlink(sem_name); } + free(new_pool->name); ovs_mutex_unlock(&init_mutex); return NULL; } @@ -267,13 +291,9 @@ ovn_fast_hmap_size_for(struct hmap *hmap, int size) /* Run a thread pool which uses a callback function to process results */ void -ovn_run_pool_callback(struct worker_pool *pool, - void *fin_result, void *result_frags, - void (*helper_func)(struct worker_pool *pool, - void *fin_result, - void *result_frags, int index)) +ovn_start_pool(struct worker_pool *pool) { - int index, completed; + int index; /* Ensure that all worker threads see the same data as the * main thread. @@ -284,8 +304,19 @@ ovn_run_pool_callback(struct worker_pool *pool, for (index = 0; index < pool->size; index++) { sem_post(pool->controls[index].fire); } +} + - completed = 0; +/* Run a thread pool which uses a callback function to process results + */ +void +ovn_complete_pool_callback(struct worker_pool *pool, + void *fin_result, void *result_frags, + void (*helper_func)(struct worker_pool *pool, + void *fin_result, + void *result_frags, int index)) +{ + int index, completed = 0; do { bool test; @@ -327,6 +358,18 @@ ovn_run_pool_callback(struct worker_pool *pool, } } while (completed < pool->size); } +/* Run a thread pool which uses a callback function to process results + */ +void +ovn_run_pool_callback(struct worker_pool *pool, + void *fin_result, void *result_frags, + void (*helper_func)(struct worker_pool *pool, + void *fin_result, + void *result_frags, int index)) +{ + start_pool(pool); + complete_pool_callback(pool, fin_result, result_frags, helper_func); +} /* Run a thread pool - basic, does not do results processing. */ @@ -373,6 +416,28 @@ ovn_fast_hmap_merge(struct hmap *dest, struct hmap *inc) inc->n = 0; } +/* Run a thread pool which gathers results in an array + * of hashes. Merge results. + */ +void +ovn_complete_pool_hash(struct worker_pool *pool, + struct hmap *result, + struct hmap *result_frags) +{ + complete_pool_callback(pool, result, result_frags, merge_hash_results); +} + +/* Run a thread pool which gathers results in an array of lists. + * Merge results. + */ +void +ovn_complete_pool_list(struct worker_pool *pool, + struct ovs_list *result, + struct ovs_list *result_frags) +{ + complete_pool_callback(pool, result, result_frags, merge_list_results); +} + /* Run a thread pool which gathers results in an array * of hashes. Merge results. */ @@ -486,7 +551,7 @@ static struct worker_control *alloc_controls(int size) static void worker_pool_hook(void *aux OVS_UNUSED) { - static struct worker_pool *pool; + struct worker_pool *pool; char sem_name[256]; /* All workers must honour the must_exit flag and check for it regularly. @@ -564,4 +629,130 @@ merge_hash_results(struct worker_pool *pool OVS_UNUSED, hmap_destroy(&res_frags[index]); } +static void +ovn_thread_pool_resize_pool(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[], void *unused OVS_UNUSED) +{ + + struct worker_pool *pool; + int value; + + if (!str_to_int(argv[2], 10, &value)) { + unixctl_command_reply_error(conn, "invalid argument"); + return; + } + + if (value > 0) { + pool_size = value; + } + LIST_FOR_EACH (pool, list_node, &worker_pools) { + if (strcmp(pool->name, argv[1]) == 0) { + resize_pool(pool, value); + unixctl_command_reply_error(conn, NULL); + } + } + unixctl_command_reply_error(conn, "pool not found"); +} + +static void +ovn_thread_pool_list_pools(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *unused OVS_UNUSED) +{ + + char *reply = NULL; + char *new_reply; + char buf[256]; + struct worker_pool *pool; + + LIST_FOR_EACH (pool, list_node, &worker_pools) { + snprintf(buf, 255, "%s : %d\n", pool->name, pool->size); + if (reply) { + new_reply = xmalloc(strlen(reply) + strlen(buf) + 1); + ovs_strlcpy(new_reply, reply, strlen(reply)); + strcat(new_reply, buf); + free(reply); + } + reply = new_reply; + } + unixctl_command_reply(conn, reply); +} + +static void +ovn_thread_pool_set_parallel_on(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[], void *unused OVS_UNUSED) +{ + int value; + bool result; + if (!str_to_int(argv[1], 10, &value)) { + unixctl_command_reply_error(conn, "invalid argument"); + return; + } + + if (!ovn_can_parallelize_hashes(true)) { + unixctl_command_reply_error(conn, "cannot enable parallel processing"); + return; + } + + if (value > 0) { + /* Change default pool size */ + ovs_mutex_lock(&init_mutex); + pool_size = value; + ovs_mutex_unlock(&init_mutex); + } + + result = ovn_set_parallel_processing(true); + unixctl_command_reply(conn, result ? "enabled" : "disabled"); +} + +static void +ovn_thread_pool_set_parallel_off(struct unixctl_conn *conn, + int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *unused OVS_UNUSED) +{ + ovn_set_parallel_processing(false); + unixctl_command_reply(conn, NULL); +} + +static void +ovn_thread_pool_parallel_status(struct unixctl_conn *conn, int argc OVS_UNUSED, + const char *argv[] OVS_UNUSED, + void *unused OVS_UNUSED) +{ + char status[256]; + + sprintf(status, "%s, default pool size %d", + get_parallel_processing() ? "active" : "inactive", + pool_size); + + unixctl_command_reply(conn, status); +} + +void +ovn_parallel_thread_pools_init(void) +{ + bool test = false; + + if (atomic_compare_exchange_strong( + &initial_pool_setup, + &test, + true)) { + ovs_mutex_lock(&init_mutex); + setup_worker_pools(false); + ovs_mutex_unlock(&init_mutex); + } + + unixctl_command_register("thread-pool/set-parallel-on", "N", 1, 1, + ovn_thread_pool_set_parallel_on, NULL); + unixctl_command_register("thread-pool/set-parallel-off", "", 0, 0, + ovn_thread_pool_set_parallel_off, NULL); + unixctl_command_register("thread-pool/status", "", 0, 0, + ovn_thread_pool_parallel_status, NULL); + unixctl_command_register("thread-pool/list", "", 0, 0, + ovn_thread_pool_list_pools, NULL); + unixctl_command_register("thread-pool/reload-pool", "Pool Threads", 2, 2, + ovn_thread_pool_resize_pool, NULL); +} + #endif diff --git a/lib/ovn-parallel-hmap.h b/lib/ovn-parallel-hmap.h index 4708f41f2..403d37736 100644 --- a/lib/ovn-parallel-hmap.h +++ b/lib/ovn-parallel-hmap.h @@ -33,6 +33,7 @@ extern "C" { #include "openvswitch/hmap.h" #include "openvswitch/thread.h" #include "ovs-atomic.h" +#include "unixctl.h" /* Process this include only if OVS does not supply parallel definitions */ @@ -93,6 +94,8 @@ struct worker_pool { sem_t *done; /* Work completion semaphorew. */ void *(*start)(void *); /* Work function. */ bool workers_must_exit; /* Pool to be destroyed flag. */ + char *name; /* Name to be used in cli commands */ + bool is_mutable; /* Can the pool be reloaded with different params */ }; /* Add a worker pool for thread function start() which expects a pointer to @@ -101,7 +104,9 @@ struct worker_pool { * size uses system defaults. */ -struct worker_pool *ovn_add_worker_pool(void *(*start)(void *), int size); +struct worker_pool *ovn_add_worker_pool(void *(*start)(void *), + int size, char *name, + bool is_mutable); /* Setting this to true will make all processing threads exit */ @@ -149,6 +154,35 @@ void ovn_run_pool_callback(struct worker_pool *pool, void *fin_result, void *fin_result, void *result_frags, int index)); +/* Start a pool. Do not wait for any results. They will be collected + * using the _complete_ functions. + */ +void ovn_start_pool(struct worker_pool *pool); + +/* Complete a pool run started using start_pool(); + * Merge results from hash frags into a final hash result. + * The hash frags must be pre-sized to the same size. + */ + +void ovn_complete_pool_hash(struct worker_pool *pool, + struct hmap *result, struct hmap *result_frags); + +/* Complete a pool run started using start_pool(); + * Merge results from list frags into a final list result. + */ + +void ovn_complete_pool_list(struct worker_pool *pool, + struct ovs_list *result, struct ovs_list *result_frags); + +/* Complete a pool run started using start_pool(); + * Call a callback function to perform processing of results. + */ + +void ovn_complete_pool_callback(struct worker_pool *pool, void *fin_result, + void *result_frags, + void (*helper_func)(struct worker_pool *pool, + void *fin_result, void *result_frags, int index)); + /* Returns the first node in 'hmap' in the bucket in which the given 'hash' * would land, or a null pointer if that bucket is empty. */ @@ -259,10 +293,16 @@ static inline void init_hash_row_locks(struct hashrow_locks *hrl) bool ovn_can_parallelize_hashes(bool force_parallel); +bool ovn_set_parallel_processing(bool enable); + +bool ovn_get_parallel_processing(void); + void ovn_destroy_pool(struct worker_pool *pool); bool ovn_resize_pool(struct worker_pool *pool, int size); +void ovn_parallel_thread_pools_init(void); + /* Use the OVN library functions for stuff which OVS has not defined * If OVS has defined these, they will still compile using the OVN * local names, but will be dropped by the linker in favour of the OVS @@ -273,9 +313,16 @@ bool ovn_resize_pool(struct worker_pool *pool, int size); #define can_parallelize_hashes(force) ovn_can_parallelize_hashes(force) +#define set_parallel_processing(enable) ovn_set_parallel_processing(enable) + +#define get_parallel_processing() ovn_get_parallel_processing() + +#define enable_parallel_processing() ovn_enable_parallel_processing() + #define stop_parallel_processing(pool) ovn_stop_parallel_processing(pool) -#define add_worker_pool(start, size) ovn_add_worker_pool(start, size) +#define add_worker_pool(start, size, name, is_mutable) \ + ovn_add_worker_pool(start, size, name, is_mutable) #define fast_hmap_size_for(hmap, size) ovn_fast_hmap_size_for(hmap, size) @@ -296,10 +343,22 @@ bool ovn_resize_pool(struct worker_pool *pool, int size); #define run_pool_callback(pool, fin_result, result_frags, helper_func) \ ovn_run_pool_callback(pool, fin_result, result_frags, helper_func) +#define start_pool(pool) ovn_start_pool(pool) + +#define complete_pool_hash(pool, result, result_frags) \ + ovn_complete_pool_hash(pool, result, result_frags) + +#define complete_pool_list(pool, result, result_frags) \ + ovn_complete_pool_list(pool, result, result_frags) + +#define complete_pool_callback(pool, fin_result, result_frags, helper_func) \ + ovn_complete_pool_callback(pool, fin_result, result_frags, helper_func) + #define destroy_pool(pool) ovn_destroy_pool(pool) #define resize_pool(pool, size) ovn_resize_pool(pool, size) +#define parallel_thread_pools_init() ovn_parallel_thread_pools_init() #ifdef __clang__ #pragma clang diagnostic pop diff --git a/northd/ovn-northd.c b/northd/ovn-northd.c index 324800c32..2f7728646 100644 --- a/northd/ovn-northd.c +++ b/northd/ovn-northd.c @@ -4358,7 +4358,6 @@ ovn_lflow_init(struct ovn_lflow *lflow, struct ovn_datapath *od, /* If this option is 'true' northd will combine logical flows that differ by * logical datapath only by creating a datapath group. */ static bool use_logical_dp_groups = false; -static bool use_parallel_build = true; static struct hashrow_locks lflow_locks; @@ -4395,7 +4394,7 @@ do_ovn_lflow_add(struct hmap *lflow_map, struct ovn_datapath *od, nullable_xstrdup(ctrl_meter), ovn_lflow_hint(stage_hint), where); hmapx_add(&lflow->od_group, od); - if (!use_parallel_build) { + if (!get_parallel_processing()) { hmap_insert(lflow_map, &lflow->hmap_node, hash); } else { hmap_insert_fast(lflow_map, &lflow->hmap_node, hash); @@ -4414,7 +4413,7 @@ ovn_lflow_add_at_with_hash(struct hmap *lflow_map, struct ovn_datapath *od, struct ovn_lflow *lflow; ovs_assert(ovn_stage_to_datapath_type(stage) == ovn_datapath_get_type(od)); - if (use_logical_dp_groups && use_parallel_build) { + if (use_logical_dp_groups && get_parallel_processing()) { lock_hash_row(&lflow_locks, hash); lflow = do_ovn_lflow_add(lflow_map, od, hash, stage, priority, match, actions, io_port, stage_hint, where, @@ -4454,7 +4453,7 @@ ovn_dp_group_add_with_reference(struct ovn_lflow *lflow_ref, return false; } - if (use_parallel_build) { + if (get_parallel_processing()) { lock_hash_row(&lflow_locks, hash); hmapx_add(&lflow_ref->od_group, od); unlock_hash_row(&lflow_locks, hash); @@ -12919,7 +12918,8 @@ static void init_lflows_thread_pool(void) { if (!pool_init_done) { - build_lflows_pool = add_worker_pool(build_lflows_thread, 0); + build_lflows_pool = add_worker_pool(build_lflows_thread, 0, + "lflows", true); pool_init_done = true; } } @@ -12951,14 +12951,11 @@ build_lswitch_and_lrouter_flows(struct hmap *datapaths, struct hmap *ports, char *svc_check_match = xasprintf("eth.dst == %s", svc_monitor_mac); - if (use_parallel_build) { + if (get_parallel_processing()) { init_lflows_thread_pool(); - if (!can_parallelize_hashes(false)) { - use_parallel_build = false; - } } - if (use_parallel_build) { + if (get_parallel_processing()) { struct hmap *lflow_segs; struct lswitch_flow_build_info *lsiv; int index; @@ -13154,7 +13151,7 @@ build_lflows(struct northd_context *ctx, struct hmap *datapaths, struct hmap lflows; fast_hmap_size_for(&lflows, max_seen_lflow_size); - if (use_parallel_build) { + if (get_parallel_processing()) { update_hashrow_locks(&lflows, &lflow_locks); } build_lswitch_and_lrouter_flows(datapaths, ports, @@ -14226,10 +14223,6 @@ ovnnb_db_run(struct northd_context *ctx, northd_probe_interval_nb = get_probe_interval(ovnnb_db, nb); northd_probe_interval_sb = get_probe_interval(ovnsb_db, nb); - use_parallel_build = - (smap_get_bool(&nb->options, "use_parallel_build", false) && - can_parallelize_hashes(false)); - use_logical_dp_groups = smap_get_bool(&nb->options, "use_logical_dp_groups", true); use_ct_inv_match = smap_get_bool(&nb->options, @@ -15144,8 +15137,9 @@ main(int argc, char *argv[]) daemonize_complete(); + ovn_parallel_thread_pools_init(); + init_hash_row_locks(&lflow_locks); - use_parallel_build = can_parallelize_hashes(false); /* We want to detect (almost) all changes to the ovn-nb db. */ struct ovsdb_idl_loop ovnnb_idl_loop = OVSDB_IDL_LOOP_INITIALIZER( diff --git a/tests/ovn-macros.at b/tests/ovn-macros.at index f06f2e68e..958ce18b0 100644 --- a/tests/ovn-macros.at +++ b/tests/ovn-macros.at @@ -179,6 +179,18 @@ ovn_start_northd() { test -d "$ovs_base/$name" || mkdir "$ovs_base/$name" as $name start_daemon $NORTHD_TYPE $northd_args -vjsonrpc \ --ovnnb-db=$OVN_NB_DB --ovnsb-db=$OVN_SB_DB + if test -z "$USE_PARALLEL_THREADS" ; then + USE_PARALLEL_THREADS=0 + fi + + if test X$NORTHD_USE_PARALLELIZATION = Xyes; then + case ${NORTHD_TYPE:=ovn-northd} in + ovn-northd) ovs-appctl --timeout=10 --target northd$suffix/ovn-northd \ + thread-pool/set-parallel-on $USE_PARALLEL_THREADS + ;; + esac + fi + } # ovn_start [--backup-northd=none|paused] [AZ] @@ -252,10 +264,6 @@ ovn_start () { else ovn-nbctl set NB_Global . options:use_logical_dp_groups=false fi - - if test X$NORTHD_USE_PARALLELIZATION = Xyes; then - ovn-nbctl set NB_Global . options:use_parallel_build=true - fi } # Interconnection networks.