[postgis-users] Slony and PostGIS problem

[Brian Peschel]

Figured I would start here since I suspect this is a PostGIS problem 
more than a Slon problem.

My main node is a Fedora 8, Postgres 8.3.1, Slon 1.2.20, and PostGIS 
1.3.  This is being replicated via slon to an Ubuntu 10, Postgres 
8.4.11, Slon 1.2.20, and PostGIS 1.4.  My table looks like this:
Table "public.ml_polygon"
  Column |   Type   | Modifiers
--------+----------+-----------
  msg_no | integer  | not null
  rev_no | integer  | not null
  geom   | geometry |
Indexes:
     "ml_polygon_pkey" PRIMARY KEY, btree (msg_no, rev_no)
     "ml_geom_idx" gist (geom)
Check constraints:
     "enforce_dims_geom" CHECK (ndims(geom) = 2)
     "enforce_srid_geom" CHECK (srid(geom) = 4269)

I have some bad geoms in the database, basically polygons with invalid 
point counts:
  70650590 |      1 | POLYGON((-90.869358 38.839444,-90.869358 
38.839447,-90.869358 38.839444))

In PostGIS 1.3, this is perfectly acceptable (even though it is bad).  
In 1.4, this doesn't work and a select gives you this error:
ERROR:  geometry requires more points

which also appears in my Postgres log.  Google tells me this was an 
issue with early PostGIS 1.4 versions.  Okay.

But I need to fix these.  So on my main node, I did this:
update ml_polygon set geom = ST_SetSRID('POLYGON((-90.869358 
38.839444,-90.869358 38.839447,-90.869358 38.839447,-90.869358 
38.839444))', 4269) where msg_no = 70650590 and rev_no = 1;

which happily updated my main node.  And nothing showed up on by second 
node.  No errors messages anywhere I could find.  New inserts in this 
table in my main node showed up on the second node, so I know slony is 
working, but this update never showed up.

I did a different update (to an existing 16 point polygon) and the 
updated showed up in my second node.  I was worried that slony wouldn't 
pick up updates of a geometry column, but this proved to me that wasn't 
the problem.  I re-did the first insert and, surprise, surprise, it 
worked fine.

So, I tried a different record with the same 3 point problem, and it 
showed up on the second node.

Needless to say, I am very confused.  Not only is the constancy of what 
works and fails a problem, but the fact the slony copy (apparently) 
failed with no error messages concerned me also.  Can I trust the data 
in my second node?

Any thoughts?
- Brian