diff options
author | Gaurav Kumar Garg <garg.gaurav52@gmail.com> | 2015-12-28 11:46:54 +0530 |
---|---|---|
committer | Atin Mukherjee <amukherj@redhat.com> | 2015-12-28 22:26:01 -0800 |
commit | c0cc93dfe6fc63caeae9448dc689adcf13ea3aae (patch) | |
tree | ca43fda901afb6ec405d87c84a553cb9841994eb | |
parent | 15965dd310192cae8f05fddda309e77ff5c72ca2 (diff) |
glusterd: reduce friend update flood
This patch is backport of: http://review.gluster.org/#/c/12999/
When in a befriended state, glusterd would broadcast friend updates to
all other peers whenver a ACC or LOCAL_ACC event occurred.
When a downed glusterd came back up and established connections again,
this lead to a flood of friend updates to happen on the order of N^2 (N
is the number of peers in the cluster)
In larger clusters this was problematic, and could lead to very long
times for the cluster to settle down when a peer came back up. Multiple
peers coming back up at the same time would compound the problem.
Broadcasting of friend updates doesn't have much use in places other
that during a peer probe. Instead of broadcasting friend updates on
connection re-establishment, updates can just be exchanged between the
peers involved in the connection.
This patch changes the glusterd friend state-machine to send updates
only to the required peer for ACC or LOCAL_ACC events when in befriended
state. The number of updates sent now is in the order of N.
For a 10 node cluster, the number of updates reduced by 5 times. When
creating the 10 node cluster, the updates reduced from ~500 to ~150.
When a glusterd restarted, the number of exchanges reduced from ~160 to
~35.
>> BUG: 1292749
>> Change-Id: Ib6072090c7069b081d018cdaa3dc878819ab1d18
>> Signed-off-by: Kaushal M <kaushal@redhat.com>
>> Reviewed-on: http://review.gluster.org/12999
>> Reviewed-by: Atin Mukherjee <amukherj@redhat.com>
>> Tested-by: NetBSD Build System <jenkins@build.gluster.org>
>> Tested-by: Gluster Build System <jenkins@build.gluster.com>
Change-Id: I389de2cc224f0ed627d98ae062209dd4f93e3b19
BUG: 1294410
Signed-off-by: Gaurav Kumar Garg <ggarg@redhat.com>
Signed-off-by: Kaushal M <kaushal@redhat.com>
Reviewed-on: http://review.gluster.org/13095
Reviewed-by: Atin Mukherjee <amukherj@redhat.com>
Tested-by: NetBSD Build System <jenkins@build.gluster.org>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
-rw-r--r-- | xlators/mgmt/glusterd/src/glusterd-sm.c | 98 |
1 files changed, 96 insertions, 2 deletions
diff --git a/xlators/mgmt/glusterd/src/glusterd-sm.c b/xlators/mgmt/glusterd/src/glusterd-sm.c index 9c772ba4d4b..4f345bc2c79 100644 --- a/xlators/mgmt/glusterd/src/glusterd-sm.c +++ b/xlators/mgmt/glusterd/src/glusterd-sm.c @@ -598,6 +598,100 @@ out: return ret; } +/* ac_update_friend only sends friend update to the friend that caused this + * event to happen + */ +static int +glusterd_ac_update_friend (glusterd_friend_sm_event_t *event, void *ctx) +{ + int ret = 0; + glusterd_peerinfo_t *cur_peerinfo = NULL; + glusterd_peerinfo_t *peerinfo = NULL; + rpc_clnt_procedure_t *proc = NULL; + xlator_t *this = NULL; + glusterd_friend_update_ctx_t ev_ctx = {{0}}; + glusterd_conf_t *priv = NULL; + dict_t *friends = NULL; + char key[100] = {0,}; + int32_t count = 0; + + GF_ASSERT (event); + + this = THIS; + priv = this->private; + + GF_ASSERT (priv); + + rcu_read_lock (); + + cur_peerinfo = glusterd_peerinfo_find (event->peerid, event->peername); + if (!cur_peerinfo) { + gf_msg (this->name, GF_LOG_ERROR, 0, + GD_MSG_PEER_NOT_FOUND, "Could not find peer %s(%s)", + event->peername, uuid_utoa (event->peerid)); + ret = -1; + goto out; + } + + /* Bail out early if peer is not connected. + * We cannot send requests to the peer until we have established our + * client connection to it. + */ + if (!cur_peerinfo->connected || !cur_peerinfo->peer) { + ret = 0; + goto out; + } + + ev_ctx.op = GD_FRIEND_UPDATE_ADD; + + friends = dict_new (); + if (!friends) + goto out; + + snprintf (key, sizeof (key), "op"); + ret = dict_set_int32 (friends, key, ev_ctx.op); + if (ret) + goto out; + + cds_list_for_each_entry_rcu (peerinfo, &priv->peers, uuid_list) { + if (!glusterd_should_update_peer (peerinfo, cur_peerinfo)) + continue; + + count++; + + memset (key, 0, sizeof (key)); + snprintf (key, sizeof (key), "friend%d", count); + ret = gd_add_friend_to_dict (peerinfo, friends, key); + if (ret) + goto out; + } + + ret = dict_set_int32 (friends, "count", count); + if (ret) + goto out; + + ret = dict_set_static_ptr (friends, "peerinfo", cur_peerinfo); + if (ret) { + gf_msg (this->name, GF_LOG_ERROR, 0, GD_MSG_DICT_SET_FAILED, + "failed to set peerinfo"); + goto out; + } + + proc = &cur_peerinfo->peer->proctable[GLUSTERD_FRIEND_UPDATE]; + if (proc->fn) + ret = proc->fn (NULL, this, friends); + + gf_msg_debug (this->name, 0, "Returning with %d", ret); + +out: + rcu_read_unlock (); + + if (friends) + dict_unref (friends); + + return ret; +} + /* Clean up stale volumes on the peer being detached. The volumes which have * bricks on other peers are stale with respect to the detached peer. */ @@ -1043,8 +1137,8 @@ glusterd_sm_t glusterd_state_befriended [] = { {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_none}, //EVENT_NONE, {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_none}, //EVENT_PROBE, {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_none}, //EVENT_INIT_FRIEND_REQ, - {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_none}, //EVENT_RCVD_ACC - {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_send_friend_update}, //EVENT_RCVD_LOCAL_ACC + {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_update_friend}, //EVENT_RCVD_ACC + {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_update_friend}, //EVENT_RCVD_LOCAL_ACC {GD_FRIEND_STATE_REJECTED, glusterd_ac_none}, //EVENT_RCVD_RJT {GD_FRIEND_STATE_REJECTED, glusterd_ac_none}, //EVENT_RCVD_LOCAL_RJT {GD_FRIEND_STATE_BEFRIENDED, glusterd_ac_handle_friend_add_req}, //EVENT_RCV_FRIEND_REQ |