A simple workaround is to make sure we always have at least one snapshot
file before upgrading (e.g. we can use a small snapCount to force snap
generation). Though from user perspective this is not ideal, but at least
this would unblock the upgrade.

I'll create a JIRA so we can discuss what's the best to address this issue.

On Mon, Jun 4, 2018 at 11:32 PM, Matteo Merli <[email protected]> wrote:

> That is correct, there are only a few transaction so the snapshot has not
> been triggered yet.
>
> The question is more on how to plan for seamless upgrade, from 3.4.10 to
> 3.5.x, from an end users perspective.
>
> On Mon, Jun 4, 2018 at 11:15 PM Michael Han <[email protected]> wrote:
>
>> Hi Matteo,
>>
>> Maybe your ZK instance did not take a snapshot at all - it's possible if
>> your total number of transactions less than the configured snapCount
>> (default value is 10000) at the time you are doing upgrade. You could check
>> your transaction log file and the snapCount configuration see if this is
>> the case or not.
>>
>>
>> On Mon, Jun 4, 2018 at 10:02 PM, Matteo Merli <[email protected]> wrote:
>>
>>>
>>> >> Also can you advice the steps for people who using 3.4.x to upgrade
>>>> to 3.5.4-beta
>>>>
>>>> The only catch I remember is that if you are using a version older than
>>>> 3.4.6, you'd need to upgrade through 3.4.6 first before upgrading to 3.5.x,
>>>> if you are doing a rolling upgrade and want to keep the liveness of the
>>>> quorum. See more https://zookeeper.apache.org/doc/r3.5.3-beta/
>>>> zookeeperReconfig.html#ch_reconfig_upgrade.
>>>>
>>>
>>>
>>> Hi Michael,
>>>
>>> This is happening upgrading from 3.4.10 to 3.5.4. It's single node
>>> embedded ZK server, part of a self-contained standalone service.
>>>
>>> Definitely , in 3.4.10 single mode, the snapshot dosesn't get created on
>>> bootstrap.
>>>
>>> Thanks,
>>> Matteo
>>> --
>>> Matteo Merli
>>> <[email protected]>
>>>
>>
>> --
> Matteo Merli
> <[email protected]>
>

Reply via email to