On Tue, Feb 15, 2011 at 11:49 AM, Heikki Linnakangas
<heikki.linnakangas@enterprisedb.com> wrote:
> On 15.02.2011 18:42, Simon Riggs wrote:
>>
>> On Sat, 2011-02-12 at 14:11 -0800, Daniel Farina wrote:
>>>
>>> This is another bit of the syncrep patch split out.
>>>
>>> I will revisit the replication timeout one Real Soon, I promise -- but
>>> I have a couple things to do today that may delay that until the
>>> evening.
>>>
>>>
>>> https://github.com/fdr/postgres/commit/ad3ce9ac62f0e128d7d1fd20d47184f867056af1
>>>
>>> Context diff supplied here.
>>
>> Greg just tipped me off to this thread a few hours ago. I saw your other
>> work on timeouts which looks good.
>>
>> I've reworked this feature myself, and its roughly the same thing you
>> have posted, so I will just add on to this thread. The major change from
>> my earlier patch is that the logic around setting xmin on the master is
>> considerably tighter, and correctly uses locking.
>
> It would be wise to also transmit the epoch in addition to xmin, to avoid
> confusion if the standby is > 2 billion transactions behind.
That case is probably hopelessly broken anyway.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company
