I have been "poking away" at bug #137
<http://www.slony.info/bugzilla/show_bug.cgi?id=137> which involves
two behavioral changes:

a) Allow the Slonik EXECUTE SCRIPT command to include a "lock="
option, allowing one to specify what tables need to be locked in order
to guard the DDL from interspersed activity.

That part wasn't terribly difficult, and seems to be working fine.

b) Rather than capturing DDL in the event, capture it in sl_log_* at
the appropriate point.

The fact that we're locking the tables in a) guards from the race
condition that there's some elapsed time between:
  1.  The moment that the DDL is submitted to alter a table, and
  2.  The moment that the DDL is captured into sl_log_*

With a small bit of struggle, I have the "capture" taking place, which
seems to be working fine.

But I'm having a non-trivial amount of "fun" trying to get the log
application to pull the relevant tuples from sl_log_*.

I had to modify the queries against sl_log_* to change part of the
condition from:

... and log_tableid in ( [list of tables] )

to:

... and ((log_tableid is null and log_cmdtype = 'S') or log_tableid in
( [list of tables] ))

That seems to construct valid queries; witness the following (slightly
hacked up) example:


slonyregress1 at localhost->  select log_origin, log_txid, log_tableid,
log_actionseq, log_cmdtype, octet_length(log_cmddata), case when
octet_length(log_cmddata) <= 4096 then log_cmddata else null end from
"_slony_regress1".sl_log_1 where log_origin = 1 and ((log_tableid is
null and log_cmdtype = 'S') or log_tableid in (1,2,3,4,5)) and
log_txid >= '771995' and log_txid < '79972045' and
"pg_catalog".txid_visible_in_snapshot(log_txid, '772045:77992045:')
union all select log_origin, log_txid, log_tableid, log_actionseq,
log_cmdtype, octet_length(log_cmddata), case when
octet_length(log_cmddata) <= 4096 then log_cmddata else null end from
"_slony_regress1".sl_log_1 where log_origin = 1 and  ((log_tableid is
null and log_cmdtype = 'S') or log_tableid in (1,2,3,4,5)) and
log_txid in (select * from
"pg_catalog".txid_snapshot_xip('771995:771995:') except select * from
"pg_catalog".txid_snapshot_xip('772045:772045:') ) order by
log_actionseq
;
 log_origin | log_txid | log_tableid | log_actionseq | log_cmdtype |
octet_length |                    case
------------+----------+-------------+---------------+-------------+--------------+---------------------------------------------
          1 |   772101 |             |            18 | S           |
        43 | alter table public.table1 drop column foo ;
          1 |   772101 |             |            19 | S           |
         1 |                                            +
            |          |             |               |             |
           |
(2 rows)

Unfortunately, the DDL is consistently seeming to be lost, anyways.

I'm speculating that there's more logic in sync_event() than is
meeting the eye, and it's checking rather harder than I'm perceiving
that "there are valid tables in this SYNC," and perhaps not bothering
to go further because even though the query can return data, it's
thinking there are no tuples to process.

Help?