Part 3 of 5 · 1 min read
System crash
The entire running process halts, losing everything in memory, but the data on disk itself survives intact.
Think of an entire store's power going out all at once, instantly stopping every single cash register mid-transaction. Every receipt that had already been physically printed and filed stays exactly as it was.
In plain words: a system crash happens when the database's running process stops entirely, due to an operating system crash, a power failure, or a hardware fault that does not damage the storage media itself. Everything that existed only in memory, such as the buffer pool's contents covered in Level 4, is lost, including any transaction that was mid-execution at the moment of the crash. However, the data already durably written to disk, including the write-ahead log, survives completely intact, since the storage media itself was never damaged.
A worked example. A sudden power failure instantly stops every in-progress transaction and loses the entire buffer pool's contents, but every log entry and data page already written to disk before the crash remains intact and readable once the system restarts.
The key lines. No new SQL; recovering from a system crash relies entirely on the durable write-ahead log, covered in depth on the next page, since nothing in memory survives to help.
System crash
A system crash is the running database process stopping entirely, losing everything held only in memory, while data already durably written to disk, including the log, survives completely intact.
Example: a store's power going out, instantly stopping every register but leaving every already-filed receipt untouched, mirrors a system crash.