Troubleshooting Database Age Increase Problem

Description

In a given database, the maximum age between the earliest and latest transactions is 2 billion (2^31). When the age of a table is greater than the value of autovacuum_freeze_max_age (400 million by default for an RDS for PostgreSQL instance), the autovacuum process freezes the table.

Once the database age exceeds 2 billion, the database breaks down and does not accept new transactions. You need to run VACUUM FULL in single-user mode to rectify the fault.

Cause Analysis

There are several possible causes:

1. The autovacuum process does not work.
2. Too much data is written to the database.
3. Temporary tables are not released for a long time.
4. There are read-only long-running transactions.

Figure 1 Cause analysis

Troubleshooting

• The autovacuum process does not work.

  There are many dead tuples in the database, and the vacuum operation is not performed. Do as follows:

  1. Check whether the autovacuum parameter is set to on.
  2. Check the value of autovacuum_freeze_max_age. The default value is 400 million for an RDS for PostgreSQL instance. If you change the value to a value greater than 1 billion, you are advised to decrease the value.
  3. Check whether the conditions for triggering autovacuum are met.

     The autovacuum_vacuum_threshold parameter specifies the minimum number of updated or deleted tuples needed to trigger a VACUUM in any one table.

     The autovacuum_vacuum_scale_factor parameter specifies a fraction of the table size to add to autovacuum_vacuum_threshold when deciding whether to trigger a VACUUM.

  4. Run the following SQL statement to check whether the autovacuum process is normal:

     select * from pg_stat_activity where backend_type like '%vacuu%';

• Too much data is written to the database.
  1. Check the database age.

     select datname, age(datfrozenxid) from pg_database where datname <> 'template1' and datname <> 'template0' order by age desc;

  2. Check whether the autovacuum parameters are properly set and compare them with those in the parameter template.

     SELECT name, setting FROM pg_settings WHERE name like '%vacuum%';

  3. Query the five oldest tables in the database.

     select relname, relfrozenxid,  age(relfrozenxid) aa from pg_class where relfrozenxid != 0 order by aa desc limit 5;

  4. Query the autovacuum status of these tables.

     SELECT schemaname, relname, last_vacuum, last_autovacuum, vacuum_count, autovacuum_count FROM pg_stat_all_tables WHERE relname='pg_toast_1335431529';

  5. Query the sizes of these tables.

     select pg_size_pretty(pg_relation_size(pg_toast_1335431536));

  6. Run the following command twice to check the value of heap_blks_scanned in the two execution results. If the value increases normally, autovacuum is running properly.

     select * from pg_stat_progress_vacuum;

     If autovacuum is running properly, check the disk read/write throughput and IOPS metrics in the last seven days. If the storage is fully occupied for a long period of time, the disk I/O is too high. Autovacuum process clearing cannot catch up with transaction ID generation and the database age increases.

     For details about the storage types and maximum throughput, see Performance Comparison of DB Instance Storage Types.

     Figure 2 Viewing disk read/write throughput
     
     Figure 3 Viewing IOPS
     
• Temporary tables are not released for a long time.
  1. For details about the troubleshooting method, see 1 to 3. If the oldest table in the query result starts with tmp_%, check whether the table is a temporary table by viewing its properties.
  2. View the properties of the oldest table. If the value of the relpersistence field is t, the table is a temporary table.

     select * from pg_class where relname ='tmp_table_pu';

     

     In a database, temporary tables are not vacuumed, but their lifecycles are not long.

     Once the connection is released, the temporary tables are reclaimed.

     Therefore, you need to check whether there are persistent connections in the database by running the following statement:

     select (now()-backend_start) duration, *from pg_stat_activity  where backend_type = 'client backend' order by duration desc nulls  last;

     When any persistent connection is found, release it and then release the temporary table. Check whether the database age decreases.

• There are long-running transactions.

  Long-running transactions are also a cause for database age increase. You can query long-running transactions of an RDS for PostgreSQL instance on the Cloud Eye console or using SQL statements.

  1. Run the following statement to check whether there are long-running transactions:

     select * from pg_stat_activity where state <> 'idle' order by xact_start;

  2. Alternatively, check the oldest active transaction duration on the Cloud Eye console and determine whether there are long-running transactions.
     

     You can only determine whether there are long-running transactions, but cannot view details about the long-running transactions on the Cloud Eye console. You are advised to use both SQL statements and Cloud Eye to identify long-running transactions.

     Figure 4 Viewing the oldest active transaction duration
     
  3. If there is a long-running transaction, run the following SQL statement to cancel the long-running transaction:

     -- Recommended:
     select pg_cancel_backend($PID);
     -- If the preceding statement is invalid, run the following statement:
     select pg_terminate_backend($PID);

  4. After canceling the long-running transaction, perform the vacuum operation on the oldest table in the database.

     vacuum "Test20231127";

  5. After the tablespace is cleared, run the following SQL statement for verification. If the value of n_dead_tup returns to 0 or is small, the restoration is complete.

     SELECT schemaname, relname, n_live_tup, n_dead_tup,
     FROM pg_stat_all_tables WHERE relname = 'Test20231127';

Solution

• The autovacuum process does not work.
  1. Check whether the autovacuum parameter is set to on. If no, set it to on and observe the database age.
  2. Check the value of autovacuum_freeze_max_age. The default value is 400 million for an RDS for PostgreSQL instance. If you change the value to a value greater than 1 billion, decrease the value and observe the database age.

• Too much data is written to the database.

  If the disk throughput reaches the performance upper limit, change the storage type.

  Run the VACUUM command to clear old tables.

• Temporary tables are not released for a long time.

  Temporary tables will not be auto-vacuumed. If the database age increases due to temporary tables, release the client connection to reclaim the temporary tables.

• There are long-running transactions.

  Cancel the long-running transactions and then run VACUUM on the oldest table in the database.

  1. Cancel the long-running transactions.

     select pg_cancel_backend($PID);

  2. Clear the table.

     vacuum table_name;

Reference

PostgreSQL official documentation