Skip to main content

WAL Trends

The WAL tab tracks the evolution of Write-Ahead Log generation, write performance and full page image activity over the selected time period. WAL trends are essential for understanding write workload intensity and planning disk, replication and backup capacity.

AWR Trends Wal

KPI Summary Cards​

Three summary cards display period-wide WAL statistics:

CardUnitDescription
Avg WAL RateMB/sAverage WAL generation rate over the selected period
Total WAL GeneratedGBTotal WAL volume produced during the period
Avg Write Timems/writeAverage time per WAL write operation

Charts​

Chart 1 — WAL Generation Rate (MB/s)​

A line chart showing the WAL generation rate in MB/s over time.

Use for:

  • Tracking write workload intensity over days, weeks or months
  • Identifying regular patterns (batch jobs, nightly ETL, end-of-month peaks)
  • Detecting unexpected WAL spikes from runaway processes or bulk operations
  • Planning WAL storage volume and replication bandwidth
  • Validating that a workload optimization reduced WAL generation
WAL and Replication

WAL must be shipped to each standby replica. A sustained high WAL rate increases replication lag risk, especially over slow or congested network links. Monitor this trend alongside replica lag metrics.

Chart 2 — WAL Records/s + Buffer Full Events​

A dual-line chart showing:

  • WAL Records/s — number of WAL records written per second
  • Buffer Full Events — number of times the WAL buffer had to be flushed early because it was full

Use for:

  • Understanding WAL write frequency vs. volume (many small records vs. few large ones)
  • Detecting WAL buffer pressure — frequent buffer full events indicate the WAL buffer is undersized
WAL Buffer Full Events

Frequent WAL buffer full events cause synchronous WAL flushes that bypass the normal write batching, increasing I/O latency for user transactions. If this trend is growing, consider increasing wal_buffers in postgresql.conf:

wal_buffers = 64MB   # default is auto-tuned to ~1/32 of shared_buffers

Chart 3 — Full Page Images (FPI/s)​

A line chart showing the rate of Full Page Images written to WAL per second.

Understanding FPI

After each checkpoint, the first modification of any data page must be written in full to WAL (not just the change delta). These are called Full Page Images (FPI). They ensure crash recovery correctness but increase WAL volume significantly.

Use for:

  • Detecting periods of high FPI rate that inflate WAL volume
  • Correlating FPI spikes with checkpoint frequency (more checkpoints = more FPIs)
  • Validating that reducing checkpoint frequency (via max_wal_size) also reduced FPI overhead
FPI and Checkpoint Frequency

FPI rate is directly proportional to checkpoint frequency. If the FPI/s trend is high and growing, check the Checkpoint & WAL Statistics section in the AWR Viewer — frequent Requested checkpoints (triggered by max_wal_size) are the likely cause.

max_wal_size = 2GB   # increase to reduce checkpoint frequency

Chart 4 — WAL Write & Sync Timing (ms)​

A dual-line chart showing the average time for WAL write and WAL sync operations in milliseconds over time.

MetricDescription
WAL Write TimeTime to write WAL records to the OS page cache
WAL Sync TimeTime to flush WAL from the OS page cache to disk (fsync)

Use for:

  • Detecting I/O latency degradation on the WAL disk over time
  • Identifying storage performance regressions after hardware or configuration changes
  • Correlating WAL sync time increases with checkpoint activity peaks
  • Validating that a storage upgrade improved WAL write performance
High WAL Sync Time

High or growing WAL sync time indicates that the storage device is struggling to keep up with fsync requests. This directly impacts transaction commit latency — every COMMIT must wait for WAL to be synced to disk (when synchronous_commit = on).

Options to investigate:

  • Move WAL to a dedicated faster disk
  • Consider synchronous_commit = off for non-critical workloads (risk of losing last few transactions on crash)
  • Use a storage device with a battery-backed write cache

Interpretation Guide​

Growing WAL Rate Without Growing TPS​

WAL volume is increasing faster than transaction count. This may indicate:

  • Queries generating more data modifications per transaction (larger updates, bulk operations)
  • Increasing FPI rate due to more frequent checkpoints
  • A new application pattern writing large rows or using bytea/JSONB columns

WAL Rate Spikes at Regular Intervals​

Regular WAL spikes typically correspond to scheduled batch jobs, nightly ETL processes or periodic autovacuum runs on high-churn tables. Verify by cross-referencing the spike timestamps with the Activity & Load tab (TPS spikes) and VACUUM Statistics in the AWR Viewer.

Sustained High WAL Sync Time​

The WAL I/O subsystem is saturated. Cross-reference with the Checkpoint & WAL Statistics section in the AWR Viewer and the Wait Events section for WALWrite or WALSync waits.

Next Steps​