Monitoring approach for Streaming Replication with Hot Standby in PostgreSQL 9.3.

The people using PostgreSQL and the Streaming Replication feature seem to ask many of the same questions:

1. How best to monitor Streaming Replication?

2. What is the best way to do that?

3. Are there alternatives, when monitoring on Standby, to using the pg_stat_replication view on Master?

4. How should I calculate replication lag-time, in seconds, minutes, etc.?

In light of these commonly asked questions, I thought a blog would help. The following are some methods I’ve found to be useful.

Monitoring is critical for large infrastructure deployments where you have Streaming Replication for:

1. Disaster recovery

2. Streaming Replication is for High Availability

3. Load balancing, when using Streaming Replication with Hot Standby

PostgreSQL has some building blocks for replication monitoring, and the following are some important functions and views which can be use for monitoring the replication:

1. pg_stat_replication view on master/primary server.

This view helps in monitoring the standby on Master. It gives you the following details:

   pid:              Process id of walsender process
   usesysid:         OID of user which is used for Streaming replication.
   usename:          Name of user which is used for Streaming replication
   application_name: Application name connected to master
   client_addr:      Address of standby/streaming replication
   client_hostname:  Hostname of standby.
   client_port:      TCP port number on which standby communicating with WAL sender
   backend_start:    Start time when SR connected to Master.
   state:            Current WAL sender state i.e streaming
   sent_location:    Last transaction location sent to standby.
   write_location:   Last transaction written on disk at standby
   flush_location:   Last transaction flush on disk at standby.
   replay_location:  Last transaction flush on disk at standby.
   sync_priority:    Priority of standby server being chosen as synchronous standby
   sync_state:       Sync State of standby (is it async or synchronous).


postgres=# select * from pg_stat_replication ;
-[ RECORD 1 ]----+---------------------------------
pid              | 1114
usesysid         | 16384
usename          | repuser
application_name | walreceiver
client_addr      |
client_hostname  |
client_port      | 52444
backend_start    | 15-MAY-14 19:54:05.535695 -04:00
state            | streaming
sent_location    | 0/290044C0
write_location   | 0/290044C0
flush_location   | 0/290044C0
replay_location  | 0/290044C0
sync_priority    | 0
sync_state       | async

2. pg_is_in_recovery() : Function which tells whether standby is still in recovery mode or not.


postgres=# select pg_is_in_recovery();
(1 row)

3. pg_last_xlog_receive_location: Function which tells location of last transaction log which was streamed by Standby and also written on standby disk.


postgres=# select pg_last_xlog_receive_location();
(1 row)

4. pg_last_xlog_replay_location: Function which tells last transaction replayed during recovery process. e.g is given below:

postgres=# select pg_last_xlog_replay_location();
(1 row)

5. pg_last_xact_replay_timestamp: This function tells about the time stamp of last transaction which was replayed during recovery. Below is an example:

postgres=# select pg_last_xact_replay_timestamp();
 15-MAY-14 20:54:27.635591 -04:00
(1 row)

Above are some important functions/views, which are already available in PostgreSQL for monitoring the streaming replication.

So, the logical next question is, “What’s the right way to monitor the Hot Standby with Streaming Replication on Standby Server?”

If you have Hot Standby with Streaming Replication, the following are the points you should monitor:

1. Check if your Hot Standby is in recovery mode or not:

For this you can use pg_is_in_recovery() function.

2.Check whether Streaming Replication is working or not.

And easy way of doing this is checking the pg_stat_replication view on Master/Primary. This view gives information only on master if Streaming Replication is working.

3. Check If Streaming Replication is not working and Hot standby is recovering from archived WAL file.

For this, either the DBA can use the PostgreSQL Log file to monitor it or utilize the following functions provided in PostgreSQL 9.3:


4. Check how far off is the Standby from Master.

There are two ways to monitor lag for Standby.

   i. Lags in Bytes: For calculating lags in bytes, users can use the pg_stat_replication view on the master with the function pg_xlog_location_diff function. Below is an example:

pg_xlog_location_diff(pg_stat_replication.sent_location, pg_stat_replication.replay_location)

which gives the lag in bytes.

  ii. Calculating lags in Seconds. The following is SQL, which most people uses to find the lag in seconds:

   SELECT CASE WHEN pg_last_xlog_receive_location() = pg_last_xlog_replay_location()
                 THEN 0
               ELSE EXTRACT (EPOCH FROM now() - pg_last_xact_replay_timestamp())
          END AS log_delay;

Including the above into your repertoire can give you good monitoring for PostgreSQL.

I will in a future post include the script that can be used for monitoring the Hot Standby with PostgreSQL streaming replication.

Write Operation: MongoDB Vs PostgreSQL 9.3 (JSON)

PostgreSQL 9.3  has lot of new improvement like the addition of new operators for JSON data type in postgreSQL, that prompted me to explore its features for NoSQL capabilities.

MongoDB is one of NoSQL solutions that have gotten a great deal of attention in the NoSQL market. So, this time I thought to do some benchmarking with the NoSQL capability of JSON in MongoDB and the JSON datatype in PostgreSQL 9.3

For this benchmark, I have used the same machine with no optimization in installation of PostgreSQL and MongoDB (since I wanted to see how things work, out of box with default installation). And I used the sample data from MongoDB’s site, around which I had developed the functions which can generate random data using the same sample for Mongo and for PostgreSQL.


In this benchmarking, I have verified following:

1. PostgreSQL COPY Vs Mongo-Import

2. Data Disk Size of PostgreSQL and Mongo for same amount of data.

3. PostgreSQL INSERT Vs Mongo Insert

Some specification before I would display the result:

1. Operating System: CentOS 6.5, 64 bit.

2. Total Memory: 1.538 GB

3. MongoDB version: 2.4.9

4. PostgreSQL: 9.3

Below is the results which I have got:

For Bulkload (COPY Vs MongoImport):

# of rows         1000      10000      100000      1000000
mongo-import (ms) 86.241679 569.761325 6940.837053 68610.69793
PG COPY (ms)      27.36344  176.705094 1769.641917 24801.23291


Disk space utilization:

# of rows        1000      10000     100000      1000000
mongo disks (mb) 208       208       208.2033236 976
pg size (mb)     0.3515625 3.2890625 32.71875    326.8984375


# of Inserts        1000        10000       100000      1000000
MONGO INSERTS (sec) 0.521397404 4.578372454 43.92753611 449.4023542
PG INSERTS (sec)    0.326254529 4.169742939 32.21799302 319.2562722


If you look at above stats, you can see PostgreSQL JSON is much better in bulk loading and INSERTs.

Best thing is that it takes less space than MongoDB and doesn’t eat up much disk space.