database is locked condition in SQLite connection layer

Goal is to understand why latency appears only under concurrent requests in SQLite’s connection layer when the database is locked condition occurs, and to identify the conditions that turn contention into observable delay versus an immediate busy error.

Constraints include the version‑sensitive pager and locking behavior, the dependence on application busy handling and transaction length, the filesystem characteristics, and the unresolved trade‑off between keeping rollback journal as the default for compatibility and enabling WAL for better concurrent read/write performance.

What factors determine whether a concurrent reader experiences a block versus an immediate SQLITE_BUSY when a writer holds an exclusive lock in rollback journal mode?

How does the zero default busy_timeout interact with application‑level retry logic to affect observed latency under load?

In what ways does WAL mode checkpointing strategy influence latency spikes that appear only under concurrent read/write workloads?