[sheepdog-users] Issue: get_vnode_info must be called in main thread

[Andrew J. Hobbs]

Valerio was mentioning this type of issue the other day.  I remain unclear if he said using less than the full cluster for strip + parity would allow failures to not halt operation.  In other words, it seems like EC is operating as if the cluster was in strict mode, halting at less than the number of nodes to accomodate the encoding scheme.

On 07/27/2014 11:23 PM, 王力锋 wrote:
Hi, All

        I create a sheepdog cluster and enable the Erasure Code feature. My cluster has 10 nodes and the erasure code is 8:2.
        However after one node of my cluster down, all the sheepdog service in other nodes are all down.
        In my opinion, the cluster should tolerate 2 nodes' failure. Do you meet the same issue?
        I enable the http service and use the zookeeper driver.

The logs are as follows:
Jul 28 11:03:43  DEBUG [main] zk_watcher(704) path:/sheepdog/member/IPv4 ip:192.168.10.11 port:7000, type:2, state:3
Jul 28 11:03:43  DEBUG [main] update_epoch_log(26) update epoch: 2, 9
Jul 28 11:03:43  DEBUG [main] for_each_object_in_path(200) /mnt/disk11, 00da5c25d0a3bbc8_0
Jul 28 11:03:43  DEBUG [main] for_each_object_in_path(200) /mnt/disk10, 00da5c254560f353_6
Jul 28 11:03:43  EMERG [main] thread_check_enter(62) PANIC: get_vnode_info must be called in main thread
Jul 28 11:03:43  EMERG [main] thread_check_enter(62) PANIC: get_vnode_info must be called in main thread
Jul 28 11:03:43  EMERG [main] crash_handler(267) sheep exits unexpectedly (Aborted).





-------------- next part --------------
A non-text attachment was scrubbed...
Name: ajhobbs.vcf
Type: text/x-vcard
Size: 353 bytes
Desc: ajhobbs.vcf
URL: <http://lists.wpkg.org/pipermail/sheepdog-users/attachments/20140728/2118b9e6/attachment-0005.vcf>