Reproducer for SERVER-54019

This adds a sharding jstest that demonstrates the inflated n / nModified on a retryable update-by-_id batch after session migration via moveChunk. The shape mirrors the reproducer attached to the ticket on 2021-01-25; modernised to ES modules and given diagnostic failure messages.

Files

• jstests/sharding/retryable_write_session_migration_inflated_n.js (128 lines)
• src/mongo/tla_plus/Sharding/TxnsMoveRange/RETRYABLE_WRITE_EXTENSION.md (74 lines)

Reproducer flow

1. {{ShardingTest( {shards: 2, mongos: 1}

   )}} with a 4-chunk layout; seed {_id: 0, x: 5, counter: 0} on shard0.

2. Open a session with retryWrites: false; drive (lsid, txnNumber=0) manually.
3. Run update with ordered: false, two update-ones {{[{q: {_id: 0}, u: {$inc: {counter: 1}

   }}, {q: {_id: 10000}, u: {$inc:

   {counter: 1}

   }}]}}; assert {{

   {n: 1, nModified: 1}

   }}.

4. moveChunk the {x: 0..10}

   chunk to shard1 with _waitForDelete: true; flush router cache.

5. Retry the same (lsid, txnNumber); assert {{ {n: 1, nModified: 1}

   }} AND counter === 1.

6. On mismatch, fail with a message naming the ticket and the inflated tuple observed vs expected.

TLA+ extension sketch

Sibling-spec design (TxnsMoveRangeWithSessionMigration.tla, not a rewrite). Adds four state variables (shardSessionTable, shardStmtResult, routerStmtSum, sessionMigrationInFlight), three new actions (ShardApplyRetryableStmt, MoveRangeWithSessionMigration, RouterAggregateRetry), two invariants (NoInflatedRetry, SessionTableCopyMonotone). Predicts a ≤6-step TLC counterexample matching the jstest trace; sibling cfg lives in the same directory.

Run

buildscripts/resmoke.py run --suites=sharding \
  jstests/sharding/retryable_write_session_migration_inflated_n.js

Related

• SERVER-54019 — retryable updates inflated n/nModified after session migration
• SPM-3190 — closed a related regime; this regime (ordered:false, batch size > 1, multiple update-ones by _id without the shard key) was not closed