Edit

HIT Lookup Strategies

CV has two strategies to lookup HITs in HitType. This is due to major performance impact of the old method we used.

The old method:

Search for 'ASSIGNABLE' HIT,

Filter HITs by their assignment: AND NOT EXISTS (SELECT 1 FROM assignment a WHERE a.hit_id = h.id AND a.worker_id = ? AND a.assignment_status IN ('SKIPPED', 'ACCEPTED', 'SUBMITTED', 'APPROVED', 'REJECTED')) AND NOT EXISTS (SELECT 1 FROM assignment a WHERE a.hit_id = h.id AND a.worker_id = ? AND a.assignment_status IN ('RETURNED', 'ABANDONED') AND a.deadline > ?) AND (SELECT COUNT(id) FROM assignment a WHERE a.hit_id = h.id AND a.assignment_status IN ('ACCEPTED', 'SUBMITTED', 'APPROVED', 'REJECTED')) < h.max_assignments

Sort by:

1. sort_order (calculated by HIT Comparator, obsolete as Comparator is used on WF side).
2. priority (obsolete but still in database)
3. number of assignments submitted for HIT. The more assignments are submitted, the more likely HIT will appear in result set.
4. Limit 20

CV is taking 20 HITs from database and takes random HIT. This is to prevent workers to preview/accept the same HIT.

Old method has been extracted as lookup strategy: AvailableHitsSlowOrderedLookupStrategy.

The new strategy — AvailableHitsFastNonOrderedLookupStrategy:

HIT has new column, prioritized, indicating that HIT has submitted assignments.

1. Query 1
   1. Search for 'ASSIGNABLE' HIT
   2. Filter HITs by their assignment (same as in slow strategy)
   3. Filter by prioritized = true
   4. Limit 20
2. UNION
3. Query 2
   1. Search for 'ASSIGNABLE' HIT
   2. Filter HITs by their assignment (same as in slow strategy)
   3. Filter by prioritized = false
   4. Limit 20
4. CV takes 40 HITs
5. Sorts result set so prioritized HITs are in the beginning of the list
6. Takes first 20 HITs from the list
7. Returns random HIT from the result of p. 6.

In case prioritized is indexed, the AvailableHitsFastNonOrderedLookupStrategy is faster than AvailableHitsSlowOrderedLookupStrategy by a factor of 10 — 1000.

To switch to AvailableHitsFastNonOrderedLookupStrategy use following JNDI setting:

<Environment name="cv/useFastHitLookupStrategy" value="true" type="java.lang.Boolean"/>