Dirty page

In InnoDB’s buffer pool, a dirty page is a 16 KB page that has been modified in memory — by an INSERT, UPDATE, or DELETE — but whose updated content has not yet been written back to the data file on disk. The on-disk version is therefore stale relative to the in-memory one.

How it works #

Every time a transaction modifies a row, InnoDB updates the corresponding page in the buffer pool and marks it as dirty. A background flush thread (page_cleaner) periodically writes dirty pages to the .ibd files using two main strategies:

• Fuzzy checkpoint: continuous, incremental flushing to keep the dirty page ratio below innodb_max_dirty_pages_pct (default 90%).
• Adaptive flushing: accelerates flushing when the redo log write rate approaches its capacity limit, preventing InnoDB from having to stall user writes to reclaim space.

-- Monitor dirty pages in real time
SELECT VARIABLE_NAME, VARIABLE_VALUE
FROM performance_schema.global_status
WHERE VARIABLE_NAME IN (
    'Innodb_buffer_pool_pages_dirty',
    'Innodb_buffer_pool_pages_total'
);

Operational context #

A stable, low dirty page count is normal. The warning sign is a sustained, monotonically increasing count: it means the flush thread cannot drain writes fast enough. Common causes include a buffer pool that is large relative to disk throughput, an innodb_io_capacity set too low, or a sudden spike in write load.

In the event of a crash, InnoDB uses the redo log to recover dirty pages that were never flushed. The more dirty pages exist at crash time, the longer crash recovery will take on restart.