Re: Snapshot synchronization, again...

On Jan7, 2011, at 12:41 , Joachim Wieland wrote:
> On Thu, Dec 30, 2010 at 7:31 AM, Joachim Wieland <joe(at)mcknight(dot)de> wrote:
> These are the implementation details and restrictions of the patch:
>
> The exporting transaction
>
> - should be read-only (to discourage people from expecting that writes of
> the exporting transaction can be seen by the importing transaction)

That seems overly strict. Wouldn't a warning about this in the docs suffice?

> - must not be a subtransaction (we don't add subxips of our own transaction
> to the snapshot, so importing the snapshot later would result in missing
> subxips)

If that is necessary, wouldn't a sub-transaction that was sub-committed before
exporting the snapshot also pose a problem? The snapshot wouldn't include
the sub-transaction's xid, so once the exporting transaction has committed any
importing transaction would suddenly see the sub-transaction's changes.

> - adds its own xid (if any) to the xip-array

Per the previous paragraph, I believe it should also add the xid's of all
non-aborted sub-transactions to the subxip-array, overflowing the array if
necessary.

> The importing transaction
>
> - will not import a snapshot of the same backend (even though it would
> probably work)
> - will not import a snapshot of a different database in the cluster
> - should be isolation level serializable
> - must not be a subtransaction (can we completely rollback on
> subxact-rollback?)

Sounds fine so far.

> - leaves curcid as is, otherwise effects of previous commands would get lost
> and magically appear later when curcid increases
> - applies xmin, xmax, xip, subxip values of the exported snapshot to
> GetActiveSnapshot() and GetTransactionSnapshot()
> - takes itself out of the xip array
> - updates MyProc->xmin but sets it only backwards but not forwards

Change a snapshot mid-transaction like that seems risky. I'd suggest to simply
decree that an importing transaction must be read-only and must not yet have set
its snapshot. So effectively, the only allowed calling sequence would be

begin;
set transaction isolation level serializable read only;
pg_import_snapshot(...);

best regards,
Florian Pflug