Handling of corrupted databases

The database offered by Tracker Miners is a fast access cache of filesystem data. In the unlikely event of major database corruption (e.g. filesystem-level corruption), this data can be disposed and reproduceably rebuilt with no major loss other than the CPU and I/O time spent.

Still, the preferred action is to attempt database repairs first. The expected order of events is:

  1. tracker-miner-fs-3 receives an unexpected TRACKER_SPARQL_ERROR_CORRUPTED error from the Tracker SPARQL library.
  2. The tracker-miner-fs-3 will handle this error by exiting uncleanly.
  3. The process gets restarted.
  4. When the database is re-opened, the Tracker SPARQL library will detect the prior corruption, and run repair attempts. See the related library documentation for more details.
  5. If the repair attempt failed, the corrupted database file(s) will be moved aside at ~/.cache/tracker3/files.$TIMESTAMP, and the filesystem data will be reindexed from scratch.