From: James Simmons <jsimmons@infradead.org>
To: Andreas Dilger <adilger@whamcloud.com>,
Oleg Drokin <green@whamcloud.com>, NeilBrown <neilb@suse.de>
Cc: Serguei Smirnov <ssmirnov@whamcloud.com>,
Lustre Development List <lustre-devel@lists.lustre.org>
Subject: [lustre-devel] [PATCH 25/39] lnet: o2iblnd: retry qp creation with reduced queue depth
Date: Thu, 21 Jan 2021 12:16:48 -0500 [thread overview]
Message-ID: <1611249422-556-26-git-send-email-jsimmons@infradead.org> (raw)
In-Reply-To: <1611249422-556-1-git-send-email-jsimmons@infradead.org>
From: Serguei Smirnov <ssmirnov@whamcloud.com>
If negotiated number of frags * queue depth is too large for
successful qp creation, reduce the queue depth in a loop
until qp creation succeeds or the queue depth dips below 2.
Remember the reduced queue depth value to use for later
connections to the same peer.
WC-bug-id: https://jira.whamcloud.com/browse/LU-12901
Lustre-commit: 8a3ef5713cc4ae ("LU-12901 o2iblnd: retry qp creation with reduced queue depth")
Signed-off-by: Serguei Smirnov <ssmirnov@whamcloud.com>
Reviewed-on: https://review.whamcloud.com/40748
Reviewed-by: Amir Shehata <ashehata@whamcloud.com>
Reviewed-by: Cyril Bordage <cbordage@whamcloud.com>
Reviewed-by: Chris Horn <chris.horn@hpe.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
Signed-off-by: James Simmons <jsimmons@infradead.org>
---
net/lnet/klnds/o2iblnd/o2iblnd.c | 33 ++++++++++++++++++++++++++-------
net/lnet/klnds/o2iblnd/o2iblnd.h | 2 ++
2 files changed, 28 insertions(+), 7 deletions(-)
diff --git a/net/lnet/klnds/o2iblnd/o2iblnd.c b/net/lnet/klnds/o2iblnd/o2iblnd.c
index 9c65524..fc515fc 100644
--- a/net/lnet/klnds/o2iblnd/o2iblnd.c
+++ b/net/lnet/klnds/o2iblnd/o2iblnd.c
@@ -336,6 +336,7 @@ int kiblnd_create_peer(struct lnet_ni *ni, struct kib_peer_ni **peerp,
peer_ni->ibp_last_alive = 0;
peer_ni->ibp_max_frags = IBLND_MAX_RDMA_FRAGS;
peer_ni->ibp_queue_depth = ni->ni_net->net_tunables.lct_peer_tx_credits;
+ peer_ni->ibp_queue_depth_mod = 0; /* try to use the default */
atomic_set(&peer_ni->ibp_refcount, 1); /* 1 ref for caller */
INIT_LIST_HEAD(&peer_ni->ibp_list);
@@ -795,13 +796,28 @@ struct kib_conn *kiblnd_create_conn(struct kib_peer_ni *peer_ni,
init_qp_attr.qp_type = IB_QPT_RC;
init_qp_attr.send_cq = cq;
init_qp_attr.recv_cq = cq;
- /* kiblnd_send_wrs() can change the connection's queue depth if
- * the maximum work requests for the device is maxed out
- */
- init_qp_attr.cap.max_send_wr = kiblnd_send_wrs(conn);
- init_qp_attr.cap.max_recv_wr = IBLND_RECV_WRS(conn);
- rc = rdma_create_qp(cmid, conn->ibc_hdev->ibh_pd, &init_qp_attr);
+ if (peer_ni->ibp_queue_depth_mod &&
+ peer_ni->ibp_queue_depth_mod < peer_ni->ibp_queue_depth) {
+ conn->ibc_queue_depth = peer_ni->ibp_queue_depth_mod;
+ CDEBUG(D_NET, "Use reduced queue depth %u (from %u)\n",
+ peer_ni->ibp_queue_depth_mod,
+ peer_ni->ibp_queue_depth);
+ }
+
+ do {
+ /* kiblnd_send_wrs() can change the connection's queue depth if
+ * the maximum work requests for the device is maxed out
+ */
+ init_qp_attr.cap.max_send_wr = kiblnd_send_wrs(conn);
+ init_qp_attr.cap.max_recv_wr = IBLND_RECV_WRS(conn);
+ rc = rdma_create_qp(cmid, conn->ibc_hdev->ibh_pd,
+ &init_qp_attr);
+ if (rc != -ENOMEM || conn->ibc_queue_depth < 2)
+ break;
+ conn->ibc_queue_depth--;
+ } while (rc);
+
if (rc) {
CERROR("Can't create QP: %d, send_wr: %d, recv_wr: %d, send_sge: %d, recv_sge: %d\n",
rc, init_qp_attr.cap.max_send_wr,
@@ -813,11 +829,14 @@ struct kib_conn *kiblnd_create_conn(struct kib_peer_ni *peer_ni,
conn->ibc_sched = sched;
- if (conn->ibc_queue_depth != peer_ni->ibp_queue_depth)
+ if (!peer_ni->ibp_queue_depth_mod &&
+ conn->ibc_queue_depth != peer_ni->ibp_queue_depth) {
CWARN("peer %s - queue depth reduced from %u to %u to allow for qp creation\n",
libcfs_nid2str(peer_ni->ibp_nid),
peer_ni->ibp_queue_depth,
conn->ibc_queue_depth);
+ peer_ni->ibp_queue_depth_mod = conn->ibc_queue_depth;
+ }
conn->ibc_rxs = kzalloc_cpt(IBLND_RX_MSGS(conn) *
sizeof(*conn->ibc_rxs),
diff --git a/net/lnet/klnds/o2iblnd/o2iblnd.h b/net/lnet/klnds/o2iblnd/o2iblnd.h
index 1fc68e1..424ca07 100644
--- a/net/lnet/klnds/o2iblnd/o2iblnd.h
+++ b/net/lnet/klnds/o2iblnd/o2iblnd.h
@@ -638,6 +638,8 @@ struct kib_peer_ni {
u16 ibp_max_frags;
/* max_peer_credits */
u16 ibp_queue_depth;
+ /* reduced value which allows conn to be created if max fails */
+ u16 ibp_queue_depth_mod;
};
extern struct kib_data kiblnd_data;
--
1.8.3.1
_______________________________________________
lustre-devel mailing list
lustre-devel@lists.lustre.org
http://lists.lustre.org/listinfo.cgi/lustre-devel-lustre.org
next prev parent reply other threads:[~2021-01-21 17:19 UTC|newest]
Thread overview: 40+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-01-21 17:16 [lustre-devel] [PATCH 00/39] lustre: update to latest OpenSFS version as of Jan 21 2021 James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 01/39] lustre: ldlm: page discard speedup James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 02/39] lustre: ptlrpc: fixes for RCU-related stalls James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 03/39] lustre: ldlm: Do not wait for lock replay sending if import dsconnected James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 04/39] lustre: ldlm: Do not hang if recovery restarted during lock replay James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 05/39] lnet: Correct handling of NETWORK_TIMEOUT status James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 06/39] lnet: Introduce constant for net ID of LNET_NID_ANY James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 07/39] lustre: ldlm: Don't re-enqueue glimpse lock on read James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 08/39] lustre: osc: prevent overflow of o_dropped James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 09/39] lustre: llite: fix client evicition with DIO James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 10/39] lustre: Use vfree_atomic instead of vfree James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 11/39] lnet: lnd: Use NETWORK_TIMEOUT for txs on ibp_tx_queue James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 12/39] lnet: lnd: Use NETWORK_TIMEOUT for some conn failures James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 13/39] lustre: llite: allow DIO with unaligned IO count James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 14/39] lustre: osc: skip 0 row for rpc_stats James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 15/39] lustre: quota: df should return projid-specific values James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 16/39] lnet: discard the callback James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 17/39] lustre: llite: try to improve mmap performance James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 18/39] lnet: Introduce lnet_recovery_limit parameter James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 19/39] lustre: mdc: avoid easize set to 0 James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 20/39] lustre: lmv: optimize dir shard revalidate James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 21/39] lustre: ldlm: osc_object_ast_clear() is called for mdc object on eviction James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 22/39] lustre: uapi: fix compatibility for LL_IOC_MDC_GETINFO James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 23/39] lustre: llite: don't check layout info for page discard James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 24/39] lustre: update version to 2.13.57 James Simmons
2021-01-21 17:16 ` James Simmons [this message]
2021-01-21 17:16 ` [lustre-devel] [PATCH 26/39] lustre: lov: fix SEEK_HOLE calcs at component end James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 27/39] lustre: lov: instantiate components layout for fallocate James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 28/39] lustre: dom: non-blocking enqueue for DOM locks James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 29/39] lustre: llite: fiemap set flags for encrypted files James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 30/39] lustre: ldlm: don't compute sumsq for pool stats James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 31/39] lustre: lov: FIEMAP support for PFL and FLR file James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 32/39] lustre: mdc: process changelogs_catalog from the oldest rec James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 33/39] lustre: ldlm: Use req_mode while lock cleanup James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 34/39] lnet: socklnd: announce deprecation of 'use_tcp_bonding' James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 35/39] lnet: o2iblnd: remove FMR-pool support James Simmons
2021-01-21 17:16 ` [lustre-devel] [PATCH 36/39] lustre: llite: return EOPNOTSUPP if fallocate is not supported James Simmons
2021-01-21 17:17 ` [lustre-devel] [PATCH 37/39] lnet: use an unbound cred in kiblnd_resolve_addr() James Simmons
2021-01-21 17:17 ` [lustre-devel] [PATCH 38/39] lustre: lov: correctly set OST obj size James Simmons
2021-01-21 17:17 ` [lustre-devel] [PATCH 39/39] lustre: cksum: add lprocfs checksum support in MDC/MDT James Simmons
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=1611249422-556-26-git-send-email-jsimmons@infradead.org \
--to=jsimmons@infradead.org \
--cc=adilger@whamcloud.com \
--cc=green@whamcloud.com \
--cc=lustre-devel@lists.lustre.org \
--cc=neilb@suse.de \
--cc=ssmirnov@whamcloud.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).