Re: PATCH: logical_work_mem and logical streaming of large in-progress transactions
Amit Kapila <amit.kapila16@gmail.com>
Commits
GET /api/v1/messages/:b64id/commits
the thread's linked commits as JSON, with link sources.
API reference →
-
Tighten the concurrent abort check during decoding.
- 2ce353fc1902 14.0 landed
-
Improve hash_create()'s API for some added robustness.
- b3817f5f7746 14.0 landed
-
Use HASH_BLOBS for xidhash.
- a1b8aa1e4eec 14.0 landed
-
Fix initialization of RelationSyncEntry for streaming transactions.
- 69bd60672af6 14.0 landed
-
Remove unused function declaration in logicalproto.h.
- ddd5f6d2609b 14.0 landed
-
Add additional tests to test streaming of in-progress transactions.
- 58b5ae9d62bd 14.0 landed
-
Fix inline marking introduced in commit 464824323e.
- ac15b499f7f9 14.0 landed
-
Add support for streaming to built-in logical replication.
- 464824323e57 14.0 landed
-
Fix the SharedFileSetUnregister API.
- 4ab77697f67a 14.0 landed
-
Fix comment in procarray.c
- 77c7267c37f7 14.0 cited
-
Suppress compiler warning in non-cassert builds.
- e942af7b8261 14.0 cited
-
Extend the BufFile interface.
- 808e13b282ef 14.0 landed
-
Mark a few logical decoding related variables with PGDLLIMPORT.
- b48cac3b10a0 14.0 landed
-
Implement streaming mode in ReorderBuffer.
- 7259736a6e5b 14.0 landed
-
Extend the logical decoding output plugin API with stream methods.
- 45fdc9738b36 14.0 landed
-
WAL Log invalidations at command end with wal_level=logical.
- c55040ccd017 14.0 landed
-
Immediately WAL-log subtransaction and top-level XID association.
- 0bead9af484c 14.0 landed
-
Allow logical replication to transfer data in binary format.
- 9de77b545313 14.0 cited
-
Only superuser can set sslcert/sslkey in postgres_fdw user mappings
- cebf9d6e6ee1 13.0 cited
-
Track statistics for spilling of changes from ReorderBuffer.
- 9290ad198b15 13.0 landed
-
Add logical_decoding_work_mem to limit ReorderBuffer memory usage.
- cec2edfa7859 13.0 landed
-
logical decoding: process ASSIGNMENT during snapshot build
- bac2fae05c77 13.0 cited
-
Emit invalidations to standby for transactions without xid.
- c6ff84b06a68 9.6.0 cited
On Thu, Jan 30, 2020 at 6:10 PM Dilip Kumar <dilipbalaut@gmail.com> wrote:
>
> On Thu, Jan 30, 2020 at 4:11 PM Amit Kapila <amit.kapila16@gmail.com> wrote:
> >
> > Also, if we need to copy the snapshot here, then do we need to again
> > copy it in ReorderBufferProcessTXN(in below code and in catch block in
> > the same function).
> I think so because as part of the
> "REORDER_BUFFER_CHANGE_INTERNAL_SNAPSHOT" change, we might directly
> point to the snapshot and that will get truncated when we truncate all
> the changes of the ReorderBufferTXN. So I think we can check if
> snapshot_now->copied is true then we can avoid copying otherwise we
> can copy?
>
Yeah, that makes sense, but I think then we also need to ensure that
ReorderBufferStreamTXN frees the snapshot only when it is copied. It
seems to me it should be always copied in the place where we are
trying to free it, so probably we should have an Assert there.
One more thing:
ReorderBufferProcessTXN()
{
..
+ if (streaming)
+ {
+ /*
+ * While streaming an in-progress transaction there is a
+ * possibility that the (sub)transaction might get aborted
+ * concurrently. In such case if the (sub)transaction has
+ * catalog update then we might decode the tuple using wrong
+ * catalog version. So for detecting the concurrent abort we
+ * set CheckXidAlive to the current (sub)transaction's xid for
+ * which this change belongs to. And, during catalog scan we
+ * can check the status of the xid and if it is aborted we will
+ * report an specific error which we can ignore. We might have
+ * already streamed some of the changes for the aborted
+ * (sub)transaction, but that is fine because when we decode the
+ * abort we will stream abort message to truncate the changes in
+ * the subscriber.
+ */
+ CheckXidAlive = change->txn->xid;
+ }
..
}
I think it is better to move the above code into an inline function
(something like SetXidAlive). It will make the code in function
ReorderBufferProcessTXN look cleaner and easier to understand.
> Other comments look fine to me so I will reply to them along with the
> next version of the patch.
>
Okay, thanks.
--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com