From mboxrd@z Thu Jan 1 00:00:00 1970 From: James Simmons Date: Wed, 15 Jul 2020 16:45:10 -0400 Subject: [lustre-devel] [PATCH 29/37] lnet: Allow router to forward to healthier NID In-Reply-To: <1594845918-29027-1-git-send-email-jsimmons@infradead.org> References: <1594845918-29027-1-git-send-email-jsimmons@infradead.org> Message-ID: <1594845918-29027-30-git-send-email-jsimmons@infradead.org> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: lustre-devel@lists.lustre.org From: Chris Horn When a final-hop router (aka edge router) is forwarding a message, if both the originator and destination of the message are mutli-rail capable, then allow the router to choose a new destination lpni if the one selected by the message originator is unhealthy or down. HPE-bug-id: LUS-8905 WC-bug-id: https://jira.whamcloud.com/browse/LU-13606 Lustre-commit: b0e8ab1a5f6f8 ("LU-13606 lnet: Allow router to forward to healthier NID") Signed-off-by: Chris Horn Reviewed-on: https://review.whamcloud.com/38798 Reviewed-by: Serguei Smirnov Reviewed-by: Amir Shehata Reviewed-by: Oleg Drokin Signed-off-by: James Simmons --- include/linux/lnet/lib-lnet.h | 4 ++-- net/lnet/lnet/lib-move.c | 37 +++++++++++++++++++++++++++++++++++-- 2 files changed, 37 insertions(+), 4 deletions(-) diff --git a/include/linux/lnet/lib-lnet.h b/include/linux/lnet/lib-lnet.h index 75c0da7..b069422 100644 --- a/include/linux/lnet/lib-lnet.h +++ b/include/linux/lnet/lib-lnet.h @@ -819,8 +819,8 @@ int lnet_get_peer_ni_info(u32 peer_index, u64 *nid, } /* - * A peer is alive if it satisfies the following two conditions: - * 1. peer health >= LNET_MAX_HEALTH_VALUE * router_sensitivity_percentage + * A peer NI is alive if it satisfies the following two conditions: + * 1. peer NI health >= LNET_MAX_HEALTH_VALUE * router_sensitivity_percentage * 2. the cached NI status received when we discover the peer is UP */ static inline bool diff --git a/net/lnet/lnet/lib-move.c b/net/lnet/lnet/lib-move.c index 2f3ef8c..234fbb5 100644 --- a/net/lnet/lnet/lib-move.c +++ b/net/lnet/lnet/lib-move.c @@ -2371,6 +2371,8 @@ struct lnet_ni * int cpt, rc; int md_cpt; u32 send_case = 0; + bool final_hop; + bool mr_forwarding_allowed; memset(&send_data, 0, sizeof(send_data)); @@ -2447,16 +2449,47 @@ struct lnet_ni * else send_case |= REMOTE_DST; + final_hop = false; + if (msg->msg_routing && (send_case & LOCAL_DST)) + final_hop = true; + + /* Determine whether to allow MR forwarding for this message. + * NB: MR forwarding is allowed if the message originator and the + * destination are both MR capable, and the destination lpni that was + * originally chosen by the originator is unhealthy or down. + * We check the MR capability of the destination further below + */ + mr_forwarding_allowed = false; + if (final_hop) { + struct lnet_peer *src_lp; + struct lnet_peer_ni *src_lpni; + + src_lpni = lnet_nid2peerni_locked(msg->msg_hdr.src_nid, + LNET_NID_ANY, cpt); + /* We don't fail the send if we hit any errors here. We'll just + * try to send it via non-multi-rail criteria + */ + if (!IS_ERR(src_lpni)) { + src_lp = lpni->lpni_peer_net->lpn_peer; + if (lnet_peer_is_multi_rail(src_lp) && + !lnet_is_peer_ni_alive(lpni)) + mr_forwarding_allowed = true; + } + CDEBUG(D_NET, "msg %p MR forwarding %s\n", msg, + mr_forwarding_allowed ? "allowed" : "not allowed"); + } + /* Deal with the peer as NMR in the following cases: * 1. the peer is NMR * 2. We're trying to recover a specific peer NI - * 3. I'm a router sending to the final destination + * 3. I'm a router sending to the final destination and MR forwarding is + * not allowed for this message (as determined above). * In this case the source of the message would've * already selected the final destination so my job * is to honor the selection. */ if (!lnet_peer_is_multi_rail(peer) || msg->msg_recovery || - (msg->msg_routing && (send_case & LOCAL_DST))) + (final_hop && !mr_forwarding_allowed)) send_case |= NMR_DST; else send_case |= MR_DST; -- 1.8.3.1