On 03/31/2015 09:19 PM, Joshua D. Drake wrote:
>
> On 03/31/2015 10:51 AM, Andres Freund wrote:
>>
>> On 2015-03-31 10:49:06 -0700, Joshua D. Drake wrote:
>>> On 03/31/2015 04:20 AM, Heikki Linnakangas wrote:
>>>> Perhaps we could consider it after a year or two, once 9.4 is indeed
>>>> very stable, but at that point you have to wonder if it's really worth
>>>> the trouble anymore. If someone has runs into that issue frequently, he
>>>> probably should just upgrade to 9.4.
>>>
>>> Ouch. That is a really poor way to look at this.
>>
>> Man.
>>
>> Easy for you to say. You're not doing the work (which would be
>> significant in this case). You're not going to be blamed if the backport
>> breaks more things than it fixed.
>
> I understand that. I am not picking on anyone. I am just saying that
> looking at the problem this way is poor, which it is. We are saying as a
> community: Your option to remove this data loss bug is to upgrade. That
> is generally not how we approach things.
Hmm, I've never considered this to be a data loss bug. I guess you can
view it that way: if you have a standby following the master, and the
master fails so that you fail over to the standby, the standby will
refuse to start up because of this, so you can't access the data.
However, the table itself is OK, it's just the index that's corrupt.
You'll need some hackery to force the system out of standby mode, but
it's not like the data has been overwritten and lost forever.
Greg Stark suggested downgrading the error to warning during recovery
mode, so that the error would not prevent you from starting up the
system. That makes a lot of sense, I think we should do that in the
back-branches.
- Heikki