Up to date access to postgres logs

Some of my Italian colleagues have made a nifty little gadget called redislog for pushing postgres logs into Redis, the distributed in-memory cache. From there it can be fed into things like logstash. I thought it would be interesting instead to make the logs available via the Redis Foreign Data Wrapper as a Postgres table. That way we would have easy access to the running logs from Postgres with almost no effort. Here’s what I did.

First I built and installed redislog and redis_fdw. Then I added redislog to my server’s shared_preload_libraries,  set log_min_duration_statement to 0 and restarted.

Then I created a database called logger and did this in it:

create extension redis_fdw;
create server localredis
  foreign data wrapper redis_fdw;
create foreign table redis_postgres_log(
    log_entry json
  server localredis
  options (tabletype 'list', singleton_key 'postgres', database '0');
create type log_type as (                                                                
  user_name text,
  database_name text,
  process_id int,
  remote_host text,
  session_id text,
  session_line_num int,
  command_tag text,
  session_start_time timestamptz,
  virtual_transaction_id text,
  transaction_id text,
  error_severity text,
  sql_state_code text,
  detail_log text,
  detail text,
  hint text,
  internal_query text,
  internal_query_pos text,
  context text,
  query text,
  query_pos text,
  file_location text,
  application_name text,
  message text,
  "@timestamp"  timestamptz);

create view postgres_log as 
  select (x).*
  from  redis_postgres_log r,
    json_populate_record(NULL::log_type, r.log_entry) as x;

alter table postgres_log
  rename "@timestamp" to log_timestamp;

After that I could select a random row from the view:

logger=# select * from public.postgres_log offset 10000 limit 1;
-[ RECORD 1 ]----------+-----------------------------------------------------------------------------------------------------
user_name              | andrew
database_name          | pgb3
process_id             | 27513
remote_host            | [local]
session_id             | 590b44c4.6b79
session_line_num       | 1936
command_tag            | UPDATE
session_start_time     | 2017-05-04 11:12:04-04
virtual_transaction_id | 3/2498
transaction_id         | 3216
error_severity         | LOG
sql_state_code         | 
detail_log             | 
detail                 | 
hint                   | 
internal_query         | 
internal_query_pos     | 
context                | 
query                  | 
query_pos              | 
file_location          | 
application_name       | pgbench
message                | duration: 0.235 ms  statement: UPDATE pgbench_tellers SET tbalance = tbalance + -4736 WHERE tid = 8;
log_timestamp          | 2017-05-04 11:12:09.593-04

There’s a bunch of work to do to make this more scalable, for example by partitioning the log. It might also be that we need to enhance redislog and/or redis_fdw to make this work better. But in principle this is a pretty nice result, a painless way of getting up to the second log entries as a postgres table.

This Post Has 2 Comments

  1. rbt says:

    Does this perform better than using the file FDW to suck in the CSV file directly from the log directory?

    • Andrew Dunstan says:

      Probably not. But I was interested to see how well it might work anyway, since redislog is also useful for other purposes. Also, I have seen cases where extraneous matter seems to have got into CSV logs making then unreadable by the File FDW. Redislog is pretty much guaranteed to have nothing but JSON.

Leave A Reply