IPC:CheckpointStart

CheckpointStart is a PostgreSQL wait event that occurs when a backend process is waiting for a checkpoint to begin.
This typically happens in scenarios where a session explicitly requests a checkpoint (e.g., via the CHECKPOINT SQL command) or when internal operations require synchronization with the start of a new checkpoint cycle.

What causes it?

Frequent Checkpoint RequestsIf your system is configured to checkpoint too often, processes may start queuing up behind the checkpointer.
- If max_wal_size is too low, the database will force a checkpoint every time that limit is hit. If you are doing bulk loads, you might be triggering checkpoints every few seconds.
- If multiple users or scripts are running the CHECKPOINT command manually at the same time, they will collide.
- if frequent backups are scheduled using pg_start_backup(), they will also trigger checkpoints.
High I/O Backpressure Before a checkpoint can "start" writing data, it must perform some administrative tasks, such as recycling WAL files and identifying "dirty" buffers in the shared_buffers.
Large shared_buffers Initialization In systems with very large shared_buffers (e.g., hundreds of gigabytes), the checkpointer has to scan the buffer metadata to determine which pages need to be written. If the CPU is under extreme pressure, this "scanning/startup" phase can take longer, causing other processes to wait for the signal that the checkpoint has officially begun.
Wait for the "Previous" Checkpoint If a checkpoint is already in progress, subsequent checkpoints must wait for the current one to finish before starting.