Обсуждение: Logical replication lag in seconds

Hello!

I currently use Slony for replication and want to switch to logical
replication. With Slony I was aware of the replication lag of each slave
in seconds. This info was available regardless if a slave was online or
offline.

For my application I need to know the "delay" of all replicas, wether
they are currently active or not. I.e. a replica may be on the other
side of the world, serving queries there, but has problems to connect to
the master. Even in this case I need to know the replication status of
the replica.

With logical replication, it seems the only available data, eve in case
a replica is offline, is in the pg_replication_slots table, for example:

-[ RECORD 1 ]-------+---------------
slot_name           | test_6
plugin              | pgoutput
slot_type           | logical
datoid              | 16402
database            | mydns
temporary           | f
active              | f
active_pid          |
xmin                |
catalog_xmin        | 1116191193
restart_lsn         | 2EA/E61E7FA8
confirmed_flush_lsn | 2EA/E626F0B0

So I guess, the replication status of this replica is the
"confirmed_flush_lsn", ie: 2EA/E626F0B0

But how to I get from 2EA/E626F0B0 to a timestamp when this transaction
was added to the WAL files?

I would appreciate any hints, or other methods to get the delay in some
time format.

thanks
Klaus

Hi Michael!

Am 21.02.2020 um 21:24 schrieb Michael Lewis:
> I am very interested in this discussion. We settled a table with a 
> single timestamp field that a script updates every minute with NOW() so 
> that we can check the timestamp of that table on the replica, assuming 
> the clocks are synced, then we will be able to compute the lag.

I have a similar workaround at the moment. But it is more a hack than a 
nice solution, ie. I also have to store the last value locally to have 
the status also available if a replica is temporarily not reachable.

Hence it would be great if the information could be retrieved from 
WAL/replication internals.

regards
Klaus

For the records - with a simple script I hacked a solution which is
purely based on the server.

1. Create a table to track the timestamp of an lsn:

CREATE TABLE lsn2date(
   lsn pg_lsn PRIMARY KEY,
   seen timestamp NOT NULL DEFAULT NOW()
);
CREATE ROLE replication_lag_user WITH LOGIN PASSWORD 'xxx';
GRANT ALL ON TABLE lsn2date TO replication_lag_user;


2. Create a script which populates the table:

# cat /etc/systemd/system/calculate_logical_replication_lag.service
[Unit]
Description=Start and auto restart service

[Install]
WantedBy=multi-user.target

[Service]
ExecStart=/usr/bin/php /path/to/calculate_logical_replication_lag.php
Restart=always
RestartSec=10


# cat calculate_logical_replication_lag.php
<?php

$dbuser="replication_lag_user";
$dbpass="xxx";
if (!$dbconn = pg_pconnect('host=127.0.0.1 dbname=mydb user='.$dbuser.'
password='.$dbpass)) {
        print "Sorry, database connection failed";
        exit;
}

$accuracy = 10; // in seconds

while (1) {
        $dbq = pg_query("INSERT INTO lsn2date (lsn) VALUES
(pg_current_wal_lsn())");
        if ($dbq === FALSE) {
                mylog(LOG_ERROR, "SQL query error: ".pg_last_error()."\n");
                exit(1);
        }

        $dbq = pg_query("DELETE FROM lsn2date WHERE lsn < (".
                          "SELECT lsn FROM lsn2date WHERE lsn < (".
                            "SELECT confirmed_flush_lsn FROM
pg_replication_slots ORDER BY confirmed_flush_lsn ASC LIMIT 1".
                          ") ORDER BY lsn DESC LIMIT 1".
                        ")"
        );
        if ($dbq === FALSE) {
                mylog(LOG_ERROR, "SQL query error: ".pg_last_error()."\n");
                exit(1);
        }
        sleep($accuracy);
}




3. Get the lag, using a function which compares the lsn of the
replication_slots with the lsn/timestamp in the lsn2date table:



CREATE OR REPLACE FUNCTION get_replication_lag() RETURNS TABLE
(subscriber name, lag bigint) AS
$BODY$
DECLARE
    subscriber name;
BEGIN
    FOR subscriber IN
        SELECT slot_name FROM pg_replication_slots
    LOOP
        RETURN QUERY SELECT slot_name, EXTRACT(EPOCH FROM
NOW()-seen)::bigint lag from lsn2date,pg_replication_slots WHERE
slot_name=subscriber AND lsn < confirmed_flush_lsn ORDER BY lsn DESC
LIMIT 1;
    END LOOP;
    RETURN;
END
$BODY$
LANGUAGE plpgsql;


# SELECT * FROM get_replication_lag() ;
 subscriber | lag
------------+-----
 reg_sjc1   |   0
 reg_ffm1   |   0
 reg_tst2   |   0
 reg_mia1   |   0
 reg_jbg1   |   0
 reg_ams1   |   0
 reg_syy1   |   0
 reg_wie1   |   0
 reg_hkg1   |   0
 reg_gnf1   |   0
 reg_tor1   |   0
 reg_sea1   |   0
 reg_chi1   |   0
 reg_dfw1   |   0
 reg_sgp1   |   0
 reg_lhr1   |   0



regards
Klaus