From nobody Fri Oct 13 23:53:57 2023 X-Original-To: dev-commits-src-all@mlmmj.nyi.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mlmmj.nyi.freebsd.org (Postfix) with ESMTP id 4S6jyZ03Kjz4wyQJ; Fri, 13 Oct 2023 23:53:58 +0000 (UTC) (envelope-from git@FreeBSD.org) Received: from mxrelay.nyi.freebsd.org (mxrelay.nyi.freebsd.org [IPv6:2610:1c1:1:606c::19:3]) (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 client-signature RSA-PSS (4096 bits) client-digest SHA256) (Client CN "mxrelay.nyi.freebsd.org", Issuer "R3" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 4S6jyY6SCwz3KVj; Fri, 13 Oct 2023 23:53:57 +0000 (UTC) (envelope-from git@FreeBSD.org) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1697241237; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding; bh=kx8EKiPIvY+h6ZyQezARijbmcR0vUSjbSw35J4oT4RM=; b=aVRj/neJljOKVx1PiSFOxL52tRdn2YT8KX3de1To+Udj7QCzAOKbpA0ahlcxwUwqexd8c8 94ZSu7T/idhDFKedZgi4qgMf0wvJP9RtPZ4IITIpb3U6fuy6r34/4ka0vJLNAtKjfSZ1bn fNAQ38EnI1TBpACYggHZJGEmB2+c6qheRNvb6di8XNGQs/p/xVAVMqupmUEf4f1+QU9Fl8 p0MBJeT5c/UvLXLXwLIrJ0NufEcymMyBlm8hsvMS3WAoP6zwqZNcdhyziN2lXuipP3ZvXY NITCTVKt55vUei2+DmeayqDh8InCgGWvtix6n6xzAanSYNi2k8uhJzvu9k6c+Q== ARC-Seal: i=1; s=dkim; d=freebsd.org; t=1697241237; a=rsa-sha256; cv=none; b=ZFrDxjF+fYve8kT2rkKhEFJvroWjjI7ZTpHT/BSg0L93Q16fQskaCVayVZ/k/ERSDGd+sc ZMGSs3r4jhna20eq1idGIhVZSxWmjVW4RuqQLy6BxMYSsJ1VF/lA4xJvXJnF0911Ascvkt 81AAWb7odt9tY8r6QZ0UVA9dfK7HBqd26wlxBXvW/OkPrVa1HvOviwQCNOKwQdSt7VEYxK 57AEWtSZ9cmnz3goRNvWtyFDT61gdhvo1mImNPVG69AkrqYyWhLXfpenPDGmVYn5fZbnwO sHpeWlF1wlpmN5z1qouklWdtefkYRbQj0rc5EXuMSR/GqMrAKZdDofXKJiybXQ== ARC-Authentication-Results: i=1; mx1.freebsd.org; none ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1697241237; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding; bh=kx8EKiPIvY+h6ZyQezARijbmcR0vUSjbSw35J4oT4RM=; b=NEibThkYMmknPj50BP+1l30dh1JzsQZVPUoVMiF6uYRKCSKT6XDAVfs1qaQf0evGTF5BIW Suflkl9iYGYRQY7MQl6ccX4hHShS71Hf1bRfsc5cwCsRuPBTTF1xoSrRTCpNN0A8957QsQ gV9K5axXVAOkTL+ejNU0v590/+TPN2l6+pXH+klRQ3XmMW2S7F+5M4Bhe18gkguF/GwNz/ XkFCwAsxfJfUa6GzmbElRyQ7oEB4KxAaFAitmlypjKBMsw0SsI1gLiOIv5Qu7z2/R7/0dU JeTISoQgJF7s9j69jhcvjjiFMDA0lAI2yTsFYh8OFda2k5A5scNDBVfpypyKZQ== Received: from gitrepo.freebsd.org (gitrepo.freebsd.org [IPv6:2610:1c1:1:6068::e6a:5]) (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) (Client did not present a certificate) by mxrelay.nyi.freebsd.org (Postfix) with ESMTPS id 4S6jyY5Xtfz1758; Fri, 13 Oct 2023 23:53:57 +0000 (UTC) (envelope-from git@FreeBSD.org) Received: from gitrepo.freebsd.org ([127.0.1.44]) by gitrepo.freebsd.org (8.17.1/8.17.1) with ESMTP id 39DNrvjs032373; Fri, 13 Oct 2023 23:53:57 GMT (envelope-from git@gitrepo.freebsd.org) Received: (from git@localhost) by gitrepo.freebsd.org (8.17.1/8.17.1/Submit) id 39DNrv9s032370; Fri, 13 Oct 2023 23:53:57 GMT (envelope-from git) Date: Fri, 13 Oct 2023 23:53:57 GMT Message-Id: <202310132353.39DNrv9s032370@gitrepo.freebsd.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-branches@FreeBSD.org From: Mateusz Guzik Subject: git: 4deb2610cc23 - stable/13 - vfs: further speed up continuous free vnode recycle List-Id: Commit messages for all branches of the src repository List-Archive: https://lists.freebsd.org/archives/dev-commits-src-all List-Help: List-Post: List-Subscribe: List-Unsubscribe: Sender: owner-dev-commits-src-all@freebsd.org X-BeenThere: dev-commits-src-all@freebsd.org MIME-Version: 1.0 Content-Type: text/plain; charset=utf-8 Content-Transfer-Encoding: 8bit X-Git-Committer: mjg X-Git-Repository: src X-Git-Refname: refs/heads/stable/13 X-Git-Reftype: branch X-Git-Commit: 4deb2610cc238fc014aace434f04ed1f22a20502 Auto-Submitted: auto-generated The branch stable/13 has been updated by mjg: URL: https://cgit.FreeBSD.org/src/commit/?id=4deb2610cc238fc014aace434f04ed1f22a20502 commit 4deb2610cc238fc014aace434f04ed1f22a20502 Author: Mateusz Guzik AuthorDate: 2023-10-11 09:42:12 +0000 Commit: Mateusz Guzik CommitDate: 2023-10-13 23:48:14 +0000 vfs: further speed up continuous free vnode recycle The primary bottleneck *was* vnode_list mtx, which got artificially worsened due to the following work done with the lock held: 1. the global heavily modified numvnodes counter was being read, inducing massive cache line ping pong 2. should the value fit limits (which it normally did) there would be an avoidable write to vn_alloc_cyclecount, which is being read outside of the lock, once more inducing traffic But if vn_alloc_cyclecount is 0, which it normally is even when facing vnode shortage, there is no need to check numvnodes nor set it to 0 again. Another problem was numvnodes adjustment (which made the locked read much worse). While it fundamentally does not scale as it is not distributed in any fashion, it was avoidably slow. When bumping over the vnode limit, it would be modified with atomics 3 times: inc + dec to backpedal in vn_alloc, then final inc in vn_alloc_hard. One can let some slop persist over calls to vnlru_free instead. In principle each thread in the system could get here and bump it, so a limit is put in place to keep things sane. Bench setup same as in prior commits: zfs, 20 separate directory trees each with 1 million files in total and 20 find(1) processes stating them in parallel (one per each tree). Total run time (in seconds) goes down as follows: vnode limit 8388608 400000 before ~20 ~35 after ~8 ~15 With this in place the primary bottleneck is now ZFS. Sponsored by: Rubicon Communications, LLC ("Netgate") (cherry picked from commit 054f45e026d898bdc8f974d33dd748937dee1d6b) --- sys/kern/vfs_subr.c | 46 +++++++++++++++++++++++++++++++--------------- 1 file changed, 31 insertions(+), 15 deletions(-) diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 54796fe6ef7d..8c831034a8e5 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -1507,6 +1507,8 @@ static u_long vnlruproc_kicks; SYSCTL_ULONG(_vfs_vnode_vnlru, OID_AUTO, kicks, CTLFLAG_RD, &vnlruproc_kicks, 0, "Number of times vnlru got woken up due to vnode shortage"); +#define VNLRU_COUNT_SLOP 100 + /* * The main freevnodes counter is only updated when a counter local to CPU * diverges from 0 by more than VNLRU_FREEVNODES_SLOP. CPUs are conditionally @@ -1649,7 +1651,8 @@ vnlru_proc_sleep(void) * * On a kernel with only stock machinery this needs anywhere between 60 and 120 * seconds to execute (time varies *wildly* between runs). With the workaround - * it consistently stays around 20 seconds. + * it consistently stays around 20 seconds [it got further down with later + * changes]. * * That is to say the entire thing needs a fundamental redesign (most notably * to accommodate faster recycling), the above only tries to get it ouf the way. @@ -1678,7 +1681,7 @@ vnlru_proc_light_pick(void) * the limit for a short period, don't bother doing anything in * that case. */ - if (rnumvnodes > desiredvnodes + 10) { + if (rnumvnodes > desiredvnodes + VNLRU_COUNT_SLOP + 10) { if (rnumvnodes - rfreevnodes >= desiredvnodes || rfreevnodes <= wantfreevnodes) { return (-1); @@ -1812,7 +1815,8 @@ vnlru_proc(void) * limit (see vn_alloc_hard), no need to call uma_reclaim if * this happens. */ - if (onumvnodes + 1000 > desiredvnodes && numvnodes <= desiredvnodes) + if (onumvnodes + VNLRU_COUNT_SLOP + 1000 > desiredvnodes && + numvnodes <= desiredvnodes) uma_reclaim(UMA_RECLAIM_DRAIN); if (done == 0) { if (force == 0 || force == 1) { @@ -1930,19 +1934,27 @@ SYSCTL_ULONG(_vfs_vnode_stats, OID_AUTO, alloc_sleeps, CTLFLAG_RD, &vn_alloc_sle "Number of times vnode allocation blocked waiting on vnlru"); static struct vnode * __noinline -vn_alloc_hard(struct mount *mp) +vn_alloc_hard(struct mount *mp, u_long rnumvnodes, bool bumped) { - u_long rnumvnodes, rfreevnodes; + u_long rfreevnodes; - mtx_lock(&vnode_list_mtx); - rnumvnodes = atomic_load_long(&numvnodes); - if (rnumvnodes + 1 < desiredvnodes) { - vn_alloc_cyclecount = 0; - mtx_unlock(&vnode_list_mtx); - goto alloc; + if (bumped) { + if (rnumvnodes > desiredvnodes + VNLRU_COUNT_SLOP) { + atomic_subtract_long(&numvnodes, 1); + bumped = false; + } } + mtx_lock(&vnode_list_mtx); + if (vn_alloc_cyclecount != 0) { + rnumvnodes = atomic_load_long(&numvnodes); + if (rnumvnodes + 1 < desiredvnodes) { + vn_alloc_cyclecount = 0; + mtx_unlock(&vnode_list_mtx); + goto alloc; + } + rfreevnodes = vnlru_read_freevnodes(); if (rfreevnodes < wantfreevnodes) { if (vn_alloc_cyclecount++ >= rfreevnodes) { @@ -1971,6 +1983,10 @@ vn_alloc_hard(struct mount *mp) /* * Wait for space for a new vnode. */ + if (bumped) { + atomic_subtract_long(&numvnodes, 1); + bumped = false; + } mtx_lock(&vnode_list_mtx); vnlru_kick_locked(); vn_alloc_sleeps++; @@ -1983,7 +1999,8 @@ vn_alloc_hard(struct mount *mp) } alloc: mtx_assert(&vnode_list_mtx, MA_NOTOWNED); - atomic_add_long(&numvnodes, 1); + if (!bumped) + atomic_add_long(&numvnodes, 1); vnlru_kick_cond(); return (uma_zalloc_smr(vnode_zone, M_WAITOK)); } @@ -1994,11 +2011,10 @@ vn_alloc(struct mount *mp) u_long rnumvnodes; if (__predict_false(vn_alloc_cyclecount != 0)) - return (vn_alloc_hard(mp)); + return (vn_alloc_hard(mp, 0, false)); rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (__predict_false(vnlru_under(rnumvnodes, vlowat))) { - atomic_subtract_long(&numvnodes, 1); - return (vn_alloc_hard(mp)); + return (vn_alloc_hard(mp, rnumvnodes, true)); } return (uma_zalloc_smr(vnode_zone, M_WAITOK));