Skip to content

Performance

APM is a single-user local application backed by SQLite. For typical studio workloads (hundreds to low thousands of jobs) no tuning is required. This page covers the available knobs if you encounter slowness.

SQLite Tuning

APM enables WAL mode (PRAGMA journal_mode = WAL) and foreign key checks on every connection. These are set in db/schema.py:get_conn().

Additional pragmas you can add to get_conn() for faster reads on spinning disks:

conn.execute("PRAGMA cache_size = -64000")   # 64 MB page cache
conn.execute("PRAGMA synchronous = NORMAL")  # safe but faster than FULL
conn.execute("PRAGMA temp_store = MEMORY")

Warning: synchronous = OFF risks corruption on power failure. Avoid it.

Full-Text Search

Migration m009 builds FTS5 virtual tables (braille_job_fts, etc.) and triggers. These keep search fast even with large job counts.

If search feels slow after bulk imports, rebuild the FTS index:

INSERT INTO braille_job_fts(braille_job_fts) VALUES('rebuild');
INSERT INTO lp_ebraille_job_fts(lp_ebraille_job_fts) VALUES('rebuild');
INSERT INTO tactile_graphics_job_fts(tactile_graphics_job_fts) VALUES('rebuild');
INSERT INTO print_job_fts(print_job_fts) VALUES('rebuild');
INSERT INTO file_object_fts(file_object_fts) VALUES('rebuild');
INSERT INTO job_metadata_fts(job_metadata_fts) VALUES('rebuild');
INSERT INTO metadata_event_fts(metadata_event_fts) VALUES('rebuild');

Pagination

All list views default to 50 rows per page. If pages feel slow, ensure that the relevant indexed columns are used. The schema defines indexes on:

  • student_id for all job tables
  • delivered for all job tables
  • (job_type, job_id) for metadata_event, job_metadata, job_file_link
  • (job_type, job_id, step_key) for job_file_link

Database Size

The file_object table stores paths, not file content. Large artifact stores do not inflate the database. Typical databases with thousands of jobs and metadata entries remain under 50 MB.

To check current size:

du -sh ~/.local/share/accessibility_mgr/accessibility_manager.db

Backup Performance

Backups use sqlite3.Connection.backup() which checkpoints the WAL and copies pages incrementally. For a 50 MB database this takes under one second. The scheduler runs in a daemon thread so it never blocks the UI.