From nobody Wed Oct 11 10:37:56 2023 X-Original-To: dev-commits-src-main@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 4S58P11B9Xz4xDK0; Wed, 11 Oct 2023 10:37:57 +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 4S58P10lWNz4MQg; Wed, 11 Oct 2023 10:37:57 +0000 (UTC) (envelope-from git@FreeBSD.org) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=freebsd.org; s=dkim; t=1697020677; 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=TyXvcOGhXh2dTL5mMRbB+U28hUjXx3dGr77gWrvxSx4=; b=Swvjn+gUyDOERiQKXRLgNwkcl5/Mta2i8nnm1OCWS5WdLNLIhHnU7xpXwI2CqsIMn3piZR Pk6WjX1qOV0JcIw5H55teraJc27HE5gTCit7eRmnDNWUIhn2BAxwIacxLn3gVLFKzDHMFV IxEgl9VjUqXsk16mUnlBKnjYZ0Y0QTpvjFdx0UXqirWB7u3KxfLiTdiHgiDxpLDb7AK3EF t8/QEWC4tk3hf6LXurt+6SeSKZA6dLu/pxwHRkMu5kDIDVOEgx0o1fvFaeBmGk+VOQXlaA 4edM0UEz+iHYZ2ohws4sizXe7y/ms4vJIKNtogyDklaff2oIniL8pWUlhEcg3w== ARC-Seal: i=1; s=dkim; d=freebsd.org; t=1697020677; a=rsa-sha256; cv=none; b=mLTp/ISJdqwl7HvKfqLheNL6vN1PU5HGJ2lShXNKeFXGwW0f+kKGT1K7WhTOiwuWAg/9LP mD/mx1Hd1xwFyQzd7iMdGsQf1aUjzzr0fF01tyHLupGe5ZoPJx1ks/l7CceZ3tyO/lK4QO xN4lxiAuN2b+7SB+kAfHqCSo9xUg50tdrN4Kyfyn9Y0/DqgkMQnx/yqUKk7z0OrJtSgEdn LR7dpRNzTZRh+2WT8ezeh+TGF7djFWWOP2c+Yw293H3yXIfENPbYaGHrJnQshsWH+8XM4a uP2vPlNI3qFfGtpImkS2vFqgayiindE2w+l1l7qchYnpFoaZdsQNh0oMhYtkRQ== 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=1697020677; 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=TyXvcOGhXh2dTL5mMRbB+U28hUjXx3dGr77gWrvxSx4=; b=X/t1GWBZElueYR92x/L3poBlIHlBZEnaEmLMMXVAiq+5lDD5c3Z+0jXZ0dePWrb2C1wdM9 wnvbXWpHZDRdKPfFgrPkaMCdumkiDg073mKC/iZSle8GQvZLATmsZX3ax+rqOS+JqcdSDx zyS3btWpnH5MZWi/QQeZjjHMfIUdq0xzuQgo6NUwHXs7iWMf18OWnAtEQeuqEVyTTlNCRI o5BiXtz0LydyMv4lJ4kTrkj+TxJwUa5vQWaUcQr1A9JkXBDCYfa27mSZT1TL7NFK4MFnA0 AmKmxytBIob3slKxVKbdKF75WJZgUACkUlH52RqtcgDuBkcwpiyqWmsZzeRq6A== 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 4S58P06q9fz3LQ; Wed, 11 Oct 2023 10:37:56 +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 39BAbu9v003015; Wed, 11 Oct 2023 10:37:56 GMT (envelope-from git@gitrepo.freebsd.org) Received: (from git@localhost) by gitrepo.freebsd.org (8.17.1/8.17.1/Submit) id 39BAbuVe003012; Wed, 11 Oct 2023 10:37:56 GMT (envelope-from git) Date: Wed, 11 Oct 2023 10:37:56 GMT Message-Id: <202310111037.39BAbuVe003012@gitrepo.freebsd.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org From: Mateusz Guzik Subject: git: 054f45e026d8 - main - vfs: further speed up continuous free vnode recycle List-Id: Commit messages for the main branch of the src repository List-Archive: https://lists.freebsd.org/archives/dev-commits-src-main List-Help: List-Post: List-Subscribe: List-Unsubscribe: Sender: owner-dev-commits-src-main@freebsd.org X-BeenThere: dev-commits-src-main@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/main X-Git-Reftype: branch X-Git-Commit: 054f45e026d898bdc8f974d33dd748937dee1d6b Auto-Submitted: auto-generated The branch main has been updated by mjg: URL: https://cgit.FreeBSD.org/src/commit/?id=054f45e026d898bdc8f974d33dd748937dee1d6b commit 054f45e026d898bdc8f974d33dd748937dee1d6b Author: Mateusz Guzik AuthorDate: 2023-10-11 09:42:12 +0000 Commit: Mateusz Guzik CommitDate: 2023-10-11 10:37:52 +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") --- 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 52c14b4a234a..99933a6b9c88 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -1490,6 +1490,8 @@ 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 @@ -1632,7 +1634,8 @@ 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. @@ -1661,7 +1664,7 @@ 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); @@ -1795,7 +1798,8 @@ 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) { @@ -1912,19 +1916,27 @@ 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) { @@ -1953,6 +1965,10 @@ 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++; @@ -1965,7 +1981,8 @@ 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)); } @@ -1976,11 +1993,10 @@ 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));