Skip to content

Lab 5 — Solutions

Step 1 — rows per shard

SELECT * FROM run_command_on_shards('fleets', 'SELECT count(*) FROM %s')
ORDER BY shardid;

Observed (your numbers may differ slightly — hashing decides):

 shardid | success | result
---------+---------+--------
  102016 | t       | 1217
  102017 | t       | 1240
  102018 | t       | 1279
  102019 | t       | 1260
  102020 | t       | 1274
  102021 | t       | 1285
  102022 | t       | 1218
  102023 | t       | 1227

Sum: 10,000 — exactly the count through the coordinator. No row lost, none duplicated.

Step 2 — where fleet 1 lives

SELECT get_shard_id_for_distribution_column('fleets', 1);
-- e.g. 102016  (shard ids differ between cluster builds — that's fine)

SELECT shardid, nodename FROM citus_shards
WHERE table_name = 'fleets'::regclass
  AND shardid = get_shard_id_for_distribution_column('fleets', 1);

Fleet 1 (Metro Cabs) lives in exactly one shard, on exactly one worker.

Interpretation

  • The coordinator's view of a distributed table is a logical union of the shards; what you query is what the shards collectively hold.
  • Hash routing is a pure function of the distribution-column value: the same fleet_id always maps to the same shard id, so a value can never be in two shards at once.