Write-Ahead Log Kernel Settings

Configuration Tuning max_wal_size vm.swappiness 60 vm.overcommit_memory 0

1GB
CheatSheet WAL size that triggers a checkpoint How aggressive kernel perfs men page swaps Assign more men to devices then actual limit
github.com/vaibhav90/PostgreSQL-Parameter-Tuning-CheatSheet
When writing > 1GB/per, set to 20GB 1 - do not swap inactive RAM pages 2 - disable process memory overcommit

Autovacuum min_wal_size 80MB Reporting & Logging dirty_ratio 40

Min. size to shrink the WAL Max # of dirty pages before writing dirty buffers
autovacuum_vacuum_threshold 50 logging_collector
Atleast 1GB in high write workload env off Set it to 15% of RAM availability
Min. # of tuple updates/deletes prior to vacuum Capture stderr output to log files
Table with > 1M rows, set to 1% # rows checkpoint_timeout 5min Switch on for csvlog and stderr dirty_backgroud_ratio 20
Max time btw auto WAL checkpoints Min. % of RAM to occupy dirty pages before flush
autovacuum_vacuum_scale_factor 0.2 20% of RAM availability
Set it to duration of ETL run log_min_duration_statement -1
# of tuple updates or deletes prior to vacuum
checkpoint_completion_target
Min exec time to start logging statements
Memory
Set it to 0, keep the configuration simple 0.5 Set to 0 to log everything
Time spent flushing dirty buffers work_mem
autovacuum_vacuum_cost_delay 2ms 4MB
0.9 to evenly distribute flushing log_destination stderr
Max. memory to be used for query workspace
Vacuum cost delay in milliseconds
Sets the destination for server log output
(max_connections)*[0.5% * RAM Size]
Set inversely proportional to I/O frequency
Asynchronous Behavior csvlog (to pipe it to pgbadger)

autovacuum_vacuum_cost_limit -1 effective_cache_size 4GB

effective_io_concurrency 1 log_directory log Tells query planner of RAM availability
Vacuum cost amt. available before napping # simultaneous requests handled by the disk Sets the destination directory for log files
25% of RAM availability (not allocated physically)
Set 1000 to increase throughput by 5x Keep separate from db & transaction log
# disks in RAID array
autovacuum_analyze_threshold 50 shared_buffer 8MB
max_worker_processes 8 log_filename %Y-%m-%d_%H%M%S.log
Min # of tuple trans. prior to analyze Exclusively dedicated to postgres for caching
Max # of concurrent worker processes Sets the file name pattern for log files
Proportional to # of autovac workers 50% of RAM availability
max_parallel_workers + other workers For auto log rotation, set weekdays/months
autovacuum_work_mem -1 maintenance_work_mem 64MB
max_parallel_workers 8 log_rotation_size 10MB
Max memory used by each worker process Memory for maintenance operations
Max. # parallel workers that can be active Auto log rotation occurs after N kilobytes
Proportional to # of autovac workers 4% of RAM availability
Alter based on # parallel queries 100MB - avoid creation of further log segments

autovacuum_max_workers 3 max_parallel_workers_per_gather 2 log_line_prefix

wal_buffers 8kB
%m
Max # of autovacuum worker processes max # of parallel processes/executor node printf-style string to begin each log line # disk page buffers in shared memory
Between 0-3 # of cores/concurrent sessions "%h:%d:%u:%c %t" On a busy, high-core machine set to 128MB