[ovs-dev] [OVN][RAFT] Follower refusing new entries from leader

Fri Dec 13 06:39:08 UTC 2019

Hi Yun,

Sorry for late reply. Thanks for trying the patch. In fact, OVSDB will
trigger compaction by itself, depending on DB size and time, and the snap
is just in the same data file. It is just in the file header (usually the
second line). So based on what you said, it seems the patch solved your
problem. Please let me know if you still see problems afterwards.

Thanks,
Han

On Mon, Dec 9, 2019 at 4:10 AM taoyunupt <taoyunupt at 126.com> wrote:

> Hi, Han,
>              I do not  encounter that problem these days  after using this
> patch.  I think there is no COMPACT in my environment.  Actaully , I don't
> see any snap file in /var/lib/openvswitch.
>
> Thanks,
> Yun
>
>
>
>
>
> At 2019-12-04 10:01:16, "Han Zhou" <hzhou at ovn.org> wrote:
>
> Hi,
>
> Could you see if this patch fixes your problem?
> https://patchwork.ozlabs.org/patch/1203951/
>
> Thanks,
> Han
>
>
> On Mon, Dec 2, 2019 at 12:28 AM Han Zhou <hzhou at ovn.org> wrote:
>
>> Sorry for the late reply. It was holiday here.
>> I didn't see such problem when there is no compaction. Did you see this
>> problem when DB compaction didn't happen? The difference is that after
>> compaction the RAFT log doesn't have any entries and all the data is in the
>> snapshot.
>>
>> On Fri, Nov 29, 2019 at 12:11 AM taoyunupt <taoyunupt at 126.com> wrote:
>>
>>> Hi,Han
>>>           Hope to receive your reply.
>>>
>>>
>>> Thanks,
>>> Yun
>>>
>>>
>>>
>>> 在 2019-11-28 16:17:07，"taoyunupt" <taoyunupt at 126.com> 写道：
>>>
>>> Hi,Han
>>>          Another question. NO COMPACT. If restart a follower , leader
>>> sender some entries during the  break time, when it has started, if it also
>>> happend to this problem?  What is the difference between simply restart and
>>> COMPACT with restart ?
>>>
>>>
>>> Thanks,
>>> Yun
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> 在 2019-11-28 13:58:36，"taoyunupt" <taoyunupt at 126.com> 写道：
>>>
>>> Hi,Han
>>>          Thanks for your reply.  I think maybe we can disconnect the
>>> failed follower from the Haproxy then synchronize the date, after all
>>> completed, reconnect it to Haproxy again. But I do not know how to
>>> synchronize actually.
>>>          It is just my naive idea. Do you have some suggestion about how
>>> to fix this problem.  If not very completed, I wii have a try.
>>>
>>>
>>> Thanks
>>> Yun
>>>
>>>
>>>
>>>
>>>
>>>
>>> 在 2019-11-28 11:47:55，"Han Zhou" <hzhou at ovn.org> 写道：
>>>
>>>
>>>
>>> On Wed, Nov 27, 2019 at 7:22 PM taoyunupt <taoyunupt at 126.com> wrote:
>>> >
>>> > Hi,
>>> >     My OVN cluster has 3 OVN-northd nodes, They are proxied by Haproxy
>>> with a VIP. Recently, I restart OVN cluster frequently.  One of the members
>>> report the logs below.
>>> >     After read the code and paper of RAFT, it seems normal process ,If
>>> the follower does not find an entry in its log with the same index and
>>> term, then it refuses the new entries.
>>> >     I think it's reasonable to refuse. But, as we could not control
>>> Haproxy or some proxy maybe, so it will happen error when an session
>>> assignate to the failed follower.
>>> >
>>> >     Does have some means or ways to solve this problem. Maybe we can
>>> kick off the failed follower or disconnect it from the haproxy then
>>> synchronize the date ?  Hope to hear your suggestion.
>>> >
>>> >
>>> > 2019-11-27T14:22:17.060Z|00240|raft|INFO|rejecting append_request
>>> because previous entry 1103,50975 not in local log (mismatch past end of
>>> log)
>>> > 2019-11-27T14:22:17.064Z|00241|raft|ERR|Dropped 34 log messages in
>>> last 12 seconds (most recently, 0 seconds ago) due to excessive rate
>>> > 2019-11-27T14:22:17.064Z|00242|raft|ERR|internal error: deferred
>>> append_reply message completed but not ready to send because message index
>>> 14890 is past last synced index 0: a2b2 append_reply "mismatch past end of
>>> log": term=1103 log_end=14891 result="inconsistency"
>>> > 2019-11-27T14:22:17.402Z|00243|raft|INFO|rejecting append_request
>>> because previous entry 1103,50975 not in local log (mismatch past end of
>>> log)
>>> >
>>> >
>>> > [root at ovn1 ~]#  ovs-appctl -t /var/run/openvswitch/ovnsb_db.ctl
>>> cluster/status OVN_Southbound
>>> > a2b2
>>> > Name: OVN_Southbound
>>> > Cluster ID: 4c54 (4c546513-77e3-4602-b211-2e200014ad79)
>>> > Server ID: a2b2 (a2b2a9c5-cf58-4724-8421-88fd5ca5d94d)
>>> > Address: tcp:10.254.8.209:6644
>>> > Status: cluster member
>>> > Role: leader
>>> > Term: 1103
>>> > Leader: self
>>> > Vote: self
>>> >
>>> > Log: [42052, 51009]
>>> > Entries not yet committed: 0
>>> > Entries not yet applied: 0
>>> > Connections: ->beaf ->9a33 <-9a33 <-beaf
>>> > Servers:
>>> >     a2b2 (a2b2 at tcp:10.254.8.209:6644) (self) next_index=15199
>>> match_index=51008
>>> >     beaf (beaf at tcp:10.254.8.208:6644) next_index=51009
>>> match_index=0
>>> >     9a33 (9a33 at tcp:10.254.8.210:6644) next_index=51009
>>> match_index=51008
>>>
>>> >
>>>
>>>
>>> I think it is a bug. I noticed that this problem happens when the
>>> cluster is restarted after DB compaction. I mentioned it in one of the test
>>> cases:
>>> https://github.com/openvswitch/ovs/blob/master/tests/ovsdb-cluster.at#L252
>>> I also mentioned another problem related to compaction:
>>> https://github.com/openvswitch/ovs/blob/master/tests/ovsdb-cluster.at#L239
>>> I was planning to debug these but didn't get the time yet. I will try to
>>> find some time next week (it would be great if you could figure it out and
>>> submit patches).
>>>
>>>
>>>
>>> Thanks,
>>> Han
>>> _______________________________________________
>>> dev mailing list
>>> dev at openvswitch.org
>>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>>>
>>