[sheepdog] epoch number mismatch

[tuji]

Hi, I've been reporting this bug

https://bugs.launchpad.net/sheepdog-project/+bug/1329806

I find what is happning sheepdog and corosync.
But I have no idea how to avoid.

for Node104
There is one "A new membership was formed" message.it is include same joined and left nodes.
And There is no "cdrv_cpg_confchg" in sheepdog.log 

-----corosync.log-------
Jul 14 10:25:26 [2880] node104 corosync notice  [TOTEM ] A new membership (10.0.1.1:206200) was formed. Members joined: 167772417 167772418 167772419 167772420 left: 167772417 167772418 167772419 167772420

-----sheepdog.log-------
none


for Node001
There is two "A new membership was formed" lines.
And there are two "cdrv_cpg_confchg" in sheepdog.log(left and joined).
 
-----corosync.log-------
Jul 14 10:25:26 [31893] node001 corosync notice  [TOTEM ] A new membership (10.0.1.1:206196) was formed. Members left: 167772520
Jul 14 10:25:26 [31893] node001 corosync notice  [TOTEM ] A new membership (10.0.1.1:206200) was formed. Members joined: 167772520

-----sheepdog.log-------
Jul 14 10:25:26  DEBUG [main] cdrv_cpg_confchg(553) mem:4, joined:0, left:1
Jul 14 10:25:26  DEBUG [main] cdrv_cpg_confchg(553) mem:5, joined:1, left:0

The Last cdrv_cpg_confchg event with joined:1 was not generated sheepdog.it was generated by corosync.
And Node001 never recive COROSYNC_MSG_TYPE_JOIN message subsequently cdrv_cpg_confchg.
So cevent->msg is not filled in __corosync_dispatch_one.

-------------------------------------------
static bool __corosync_dispatch_one(struct corosync_event *cevent)
{
        struct sd_node entries[SD_MAX_NODES], *node;
        struct cpg_node *n;
        int idx;

        switch (cevent->type) {
        case COROSYNC_EVENT_TYPE_JOIN:
                if (!cevent->msg)
                        /* we haven't receive JOIN yet */
                        return false;
-------------------------------------------



--------------------------
Masahiro Tsuji

A.T.WORKS, INC
URL http://www.atworks.co.jp