On 10.01.2011 16:49, Simon Riggs wrote:
> On Mon, 2011-01-10 at 15:20 +0100, Magnus Hagander wrote:
>> On Sun, Jan 9, 2011 at 15:53, Simon Riggs<simon@2ndquadrant.com> wrote:
>>> On Sun, 2011-01-09 at 12:52 +0100, Magnus Hagander wrote:
>>>
>>>> One thing I noticed is that it gives an interesting property to my
>>>> patch for streaming base backups - they now show up in
>>>> pg_stat_replication, with a streaming location of 0/0.
>>>>
>>>> If the view is named pg_stat_replication, we probably want to filter
>>>> that out somehow. But then, do we want a separate view listing the
>>>> walsenders that are busy sending base backups?
>>>>
>>>> For that matter, do we want an indication that separates a walsender
>>>> not sending data from one sending that happens to be at location 0/0?
>>>> Most will leave 0/0 really quickly, but a walsender can be idle (not
>>>> received a command yet), or it can be running IDENTIFY_SYSTEM for
>>>> example.
>>>
>>> I think we need a status enum. ('BACKUP', 'CATCHUP', 'STREAM') for the 3
>>> phases of replication.
>>
>> That seems reasonable. But if we keep BACKUP in there, should we
>> really have it called pg_stat_replication? (yeah, I know, I'm not
>> giving up :P)
>>
>> (You'd need a 4th mode for WAITING or so, to indicate it's waiting for
>> a command)
>
> That's something different.
>
> The 3 phases are more concrete.
>
> BACKUP --> CATCHUP<---> STREAM
>
> When you connect you either do BACKUP or CATCHUP. Once in CATCHUP mode
> you never issue a BACKUP. Once we have caught up we move to STREAM. That
> has nothing to do with idle/active.
So how does a walsender that's waiting for a command from the client
show up? Surely it's not in "catchup" mode yet?
-- Heikki Linnakangas EnterpriseDB http://www.enterprisedb.com
