GetSnapshotData() comments

Поиск
Список
Период
Сортировка
От Robert Haas
Тема GetSnapshotData() comments
Дата
Msg-id CA+Tgmoa+90P5UcFBPPAi+DJr3vHksinek+dCqVu=MCePOuvLiA@mail.gmail.com
обсуждение исходный текст
Ответы Re: GetSnapshotData() comments  (Bruce Momjian <bruce@momjian.us>)
Список pgsql-hackers
I think that the first sentence, in the following comment within
GetSnapshotData(), is not quite right, because at the time this is
executed, we already hold ProcArrayLock, which is the only lock that
gets grabbed here:
       /*        * If we're in recovery then snapshot data comes from a different place,        * so decide which route
wetake before grab the lock. It is
 
possible for        * recovery to end before we finish taking snapshot, and for newly        * assigned transaction ids
tobe added to the procarray. Xmax cannot        * change while we hold ProcArrayLock, so those newly added transaction
     * ids would be filtered away, so we need not be concerned about them.        */
snapshot->takenDuringRecovery= RecoveryInProgress();
 

Having thought it over for a bit, I believe that the code is correct
and it's only the comment that needs to be updated.  If we were to set
snapshot->takenDuringRecovery before acquiring ProcArrayLock, then the
following sequence of events might occur:

T1: Enter GetSnapshotData().  Set snapshot->takenDuringRecovery = true.
Recovery ends
T2: Begin, get an XID.
T3: Begin, get an XID.
T3: Commit, advancing latestCompletedXid.
T1: Acquire ProcArrayLock and use the "in recovery" path, missing T2's
XID (because it's not in the subxip[] array).
T1: Do some stuff with the snapshot... not seeing T2's XID...
T2: Commit
T1: Do some stuff with the snapshot... now seeing T2's XID...

I think if we just delete the first sentence of the comment, the rest
is all correct.

A little further down, there is this comment:
               /*                * Spin over procArray checking xid, xmin, and
subxids.  The goal is                * to gather all active xids, find the lowest xmin,
and try to record                * subxids. During recovery no xids will be assigned,
so all normal                * backends can be ignored, nor are there any VACUUMs
running. All                * prepared transaction xids are held in
KnownAssignedXids, so these                * will be seen without needing to loop through procs here.
*/

...but this code is only executed when recovery is not in progress.
So I feel like everything after "try to record subxids" should either
be removed, or relocated to the following "else" block, where the
recovery path is discussed in detail.

-- 
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company


В списке pgsql-hackers по дате отправления:

Предыдущее
От: Andrew Dunstan
Дата:
Сообщение: Re: plperl crash with Debian 6 (64 bit), pl/perlu, libwww and https
Следующее
От: Tom Lane
Дата:
Сообщение: Re: error: could not find pg_class tuple for index 2662