Streaming Replication in PostgreSQL9.0 b4: Standby dropping connections

From: Chip Steele <steele(dot)chip(at)gmail(dot)com>
To: pgsql-novice(at)postgresql(dot)org
Subject: Streaming Replication in PostgreSQL9.0 b4: Standby dropping connections
Date: 2010-08-20 20:15:32
Message-ID: AANLkTimtYNphFQ0pAHT5oZjCzn9eYG+Bfv2=75_t-Pkp@mail.gmail.com
Views: Raw Message | Whole Thread | Download mbox | Resend email
Thread:
Lists: pgsql-novice

Hi,

--- Reposting with more information ---
In PG9.0-b4 I have successfully set up a primary-secondary pair with
streaming replication, with the secondary in hot_standby mode.

I have an application on the standby (and primary) which has constant
connections open to the database. Occasionally it refreshes its cache of
data from the hot_standby secondary server.
I am intermittently getting and error when updating the application's cache.

The cache update involves 2 queries:
SELECT COUNT(*) FROM core_list; -- this table has about 5 Million records
and
SELECT line, COUNT(*) FROM core_state WHERE state='normal' GROUP BY line
ORDER BY line; -- this table has about 50million records

normal operation of the database would not update the core_list table, and
would update maybe up to 10% of the core_state table between application
cache updates

table definitions follow:
CREATE TABLE core_list(
id int NOT NULL DEFAULT nextval('core_list_id_seq'),
name varchar(64) NOT NULL,
valid BOOLEAN NOT NULL DEFAULT '1'
);

unique index on the first 2 fields, index on the 3rd.

CREATE TABLE core_state(
core_id int NOT NULL,
line smallint NOT NULL,
state core_state NOT NULL
);

CREATE INDEX cs_core_id_idx ON core_state(core_id);
CREATE INDEX cs_state_line_idx ON core_state(state,line);
CREATE UNIQUE INDEX cs_core_id_line_idx ON core_state(core_id,line)

When I check the PostgreSQL logs with the same timestamp, I see the
following messages:

FATAL: terminating connection due to conflict with recovery
DETAIL: User query might have needed to see row versions that must be
removed.
HINT: In a moment you should be able to reconnect to the database and repeat
your command.

Now, this message explains that there is a problem, but I couldn't find
anything in the streaming replication/hot standby documentation indicating
that this sort of thing might occur.
Is there any way to avoid it? Any further information about this specific
problem? Can I provide anything else?

Thanks in advance,

Chip

Browse pgsql-novice by date

  From Date Subject
Next Message Frank Bax 2010-08-23 11:24:42 create a box from point?
Previous Message Rikard Bosnjakovic 2010-08-20 16:05:49 Re: Easiest way to extract owner-id from a third table