SERVER-48198 Account for extended range bounds when recovering migration decision

(cherry picked from commit 9d8eb69d583b89682520ec58595e558d5f6cc9a2)

1 files changed, 63 insertions, 0 deletions

diff --git a/jstests/sharding/range_deleter_interacts_correctly_with_refine_shard_key.js b/jstests/sharding/range_deleter_interacts_correctly_with_refine_shard_key.js
index d5139a88ce2..def5e94287c 100644
--- a/jstests/sharding/range_deleter_interacts_correctly_with_refine_shard_key.js
+++ b/jstests/sharding/range_deleter_interacts_correctly_with_refine_shard_key.js
@@ -179,6 +179,69 @@ function test(st, description, testBody) {
              hangDonorAtEndOfMigration.off();
          });
 
+    test(st,
+         "Migration recovery recovers correct decision for migration committed before shard key " +
+             "refine",
+         () => {
+             // Enable a failpoint that makes the migration donor hang before making a decision and
+             // begin a migration that hits this failpoint.
+             let hangBeforeWritingDecisionFailpoint =
+                 configureFailPoint(st.rs0.getPrimary(), "hangBeforeMakingCommitDecisionDurable");
+             const parallelMoveChunk = startParallelShell(
+                 funWithArgs(function(ns, shardKeyValueInChunk, toShardName) {
+                     assert.commandFailedWithCode(
+                         db.adminCommand(
+                             {moveChunk: ns, find: shardKeyValueInChunk, to: toShardName}),
+                         ErrorCodes.InterruptedDueToReplStateChange);
+                 }, ns, shardKeyValueInChunk, st.shard1.shardName), st.s.port);
+
+             jsTestLog("Waiting for the migration to hang before writing a decision");
+             hangBeforeWritingDecisionFailpoint.wait();
+
+             // Step up a new primary, which will interrupt the migration and trigger the migration
+             // recovery process. Set a failpoint on the new primary that will pause the recovery
+             // before it can load the latest metadata.
+             jsTestLog("Stepping up a new primary");
+             const newPrimary = st.rs0.getSecondary();
+             let hangInMigrationRecoveryFailpoint =
+                 configureFailPoint(newPrimary, "hangBeforeFilteringMetadataRefresh");
+             assert.commandWorked(newPrimary.adminCommand({replSetStepUp: 1}));
+             st.rs0.waitForState(newPrimary, ReplSetTest.State.PRIMARY);
+             st.rs0.awaitNodesAgreeOnPrimary();
+
+             jsTestLog("Waiting for the new primary to hang in migration recovery");
+             hangInMigrationRecoveryFailpoint.wait();
+
+             // Clean up the failpoint on the old primary.
+             hangBeforeWritingDecisionFailpoint.off();
+
+             // Refine the collection's shard key while the recovery task is hung.
+             jsTestLog("Refining the shard key");
+             assert.commandWorked(st.s.getCollection(ns).createIndex({x: 1, y: 1, z: 1}));
+             assert.commandWorked(
+                 st.s.adminCommand({refineCollectionShardKey: ns, key: {x: 1, y: 1, z: 1}}));
+
+             // Allow the recovery to continue by disabling the failpoint and verify that despite
+             // the recovered migration having fewer fields in its bounds than in the current shard
+             // key, the decision should be recovered successfully and orphans should be removed
+             // from the donor.
+             jsTestLog("Waiting for orphans to be removed from shard 0");
+             hangInMigrationRecoveryFailpoint.off();
+             assert.soon(() => {
+                 return st.rs0.getPrimary().getCollection(ns).find().itcount() == 0;
+             });
+
+             // Verify we can move the chunk back to the original donor once the orphans are gone.
+             awaitRSClientHosts(
+                 st.rs1.getPrimary(), st.rs0.getPrimary(), {ok: true, ismaster: true});
+             assert.commandWorked(st.s.adminCommand({
+                 moveChunk: ns,
+                 find: {x: 1, y: 1, z: 1},
+                 to: st.shard0.shardName,
+                 _waitForDelete: true
+             }));
+         });
+
     // This test was created to reproduce a specific bug, which is why it may sound like an odd
     // thing to test. See SERVER-46386 for more details.
     test(st,