Skip to content

test_replication_upgrade_delta_sync.md

Latest commit

 

History

History
129 lines (109 loc) · 6.22 KB

test_replication_upgrade_delta_sync.md

File metadata and controls

129 lines (109 loc) · 6.22 KB

Test Cases

These tests cover delta-sync replication behavior across a simulated 3.x → 4.x SGW upgrade. The pre-upgrade state is materialised by restoring the prebuilt upgrade CBS backup (revtree-only docs, no HLV xattrs) and resetting the CBL local DB from the matching upgrade cblite2 file; both binaries run 4.x throughout. Delta sync is enabled on the SGW upgrade database.

#1 test_delta_sync_history_pull_post_upgrade_sgw_mutation

Description

PULL replication of a doc whose 2nd revision was created on 4.x SGW (so SGW holds both a revtree and an HLV) by a client that still holds the revtree-only ancestor. With delta sync enabled, SGW must populate the rev message's history field with the revtree predecessors so the client can ingest the delta. The current build sends an empty history here — this test is the regression marker for the fix, and is marked xfail(strict=True) until the SGW fix lands.

Uses doc nonconflict_3 from the prebuilt upgrade dataset. The 2nd revision is created in-test by mutating the doc on 4.x SGW, which adds an HLV in parallel with the new revtree leaf (4.x SGW always writes both).

+------------------+-------------------------------+-------------------------------+
|                  |             CBL               |              SGW              |
|                  +---------------+---------------+---------------+---------------+
|                  |   Rev Tree    |      HLV      |   Rev Tree    |      HLV      |
+------------------+---------------+---------------+---------------+---------------+
| After restore    |     2-abc     |     none      |     2-abc     |     none      |
| After SGW mutate |     2-abc     |     none      | 3-xxx, 2-abc  |   [N@SGW]     |
| Expected post-PULL|     none     |    [N@SGW]    | 3-xxx, 2-abc  |   [N@SGW]     |
+------------------+---------------+---------------+---------------+---------------+

Steps

  1. Delete Sync Gateway 'upgrade' database if exists.
  2. Restore Couchbase Server Bucket using upgrade dataset.
  3. Wait 2s to ensure SG picks up the restored database.
  4. Reset local database, and load upgrade dataset.
  5. Create SG 'upgrade' database with delta_sync enabled and import from bucket. On 412 (already exists), force-recreate by delete_database + put_database.
  6. Verify delta_sync is actually enabled on SGW 'upgrade' database by fetching the live config; fail with the active config dumped if not enabled.
  7. Create user user1 with full access to _default._default.
  8. Mutate nonconflict_3 on 4.x SGW to produce a new revtree leaf + fresh HLV.
  9. Start a replicator:
    • endpoint: /upgrade
    • collections: _default._default
    • type: pull
    • document_ids: ['nonconflict_3']
    • continuous: False
    • credentials: user1/pass
  10. Wait until the replicator is stopped.
  11. Validate revid and HLV of local and remote doc:
    • Pre: local has revid + no HLV; SGW has revid + canonical HLV (not RTE-encoded).
    • Post: local has no revid (4.x CBL is HLV-only); local HLV equals SGW HLV.

Expected Outcome

✅ With SGW fix: CBL ingests the delta, ends up HLV-only with HLV matching SGW. ❌ Without SGW fix (current build): rev message's history field is empty, client cannot ingest → test fails the postcondition. The xfail(strict=True) marker makes this an expected failure today and an XPASS (CI failure) the day the fix lands, forcing the developer landing the fix to remove the marker.

#2 test_delta_sync_history_pull_pre_upgrade_sgw_two_revs

Description

PULL replication of a doc whose 2nd revision was created on 3.x SGW (so both sides have revtree-only state, no HLV anywhere) by a client that holds the revtree-only ancestor. With delta sync enabled, the client must pull the newer revtree-only rev and generate an HLV locally using the Revision-Tree-Encoding (RTE) format. This case is expected to work on the current build (via the pre-fix code path) and serves as a forward regression marker once the SGW fix lands.

Uses doc nonconflict_2 from the prebuilt upgrade dataset — its baked-in state already matches the required pre-state, so no in-test SGW mutation is needed.

+------------------+-------------------------------+-------------------------------+
|                  |             CBL               |              SGW              |
|                  +---------------+---------------+---------------+---------------+
|                  |   Rev Tree    |      HLV      |   Rev Tree    |      HLV      |
+------------------+---------------+---------------+---------------+---------------+
| After restore    |    1-abc      |     none      | 2-def, 1-abc  |     none      |
| Expected post-PULL|    none      | [2def@RTE]    | 2-def, 1-abc  |     none      |
+------------------+---------------+---------------+---------------+---------------+

Steps

  1. Delete Sync Gateway 'upgrade' database if exists.
  2. Restore Couchbase Server Bucket using upgrade dataset.
  3. Wait 2s to ensure SG picks up the restored database.
  4. Reset local database, and load upgrade dataset.
  5. Create SG 'upgrade' database with delta_sync enabled and import from bucket. On 412 (already exists), force-recreate by delete_database + put_database.
  6. Verify delta_sync is actually enabled on SGW 'upgrade' database by fetching the live config; fail with the active config dumped if not enabled.
  7. Create user user1 with full access to _default._default.
  8. Start a replicator:
    • endpoint: /upgrade
    • collections: _default._default
    • type: pull
    • document_ids: ['nonconflict_2']
    • continuous: False
    • credentials: user1/pass
  9. Wait until the replicator is stopped.
  10. Validate revid and HLV of local and remote doc:
    • Pre: both sides have revid and no HLV; CBL revid < SGW revid.
    • Post: local has no revid (4.x CBL is HLV-only); local HLV ends with @Revision+Tree+Encoding; SGW HLV is still None (PULL doesn't touch SGW).

Expected Outcome

✅ Today (pre-fix): the existing code path correctly pulls the revtree-only rev and the client generates an RTE-encoded HLV. Test passes. ✅ After fix lands: same path, same outcome. Test continues to pass. This is the forward regression marker confirming the fix doesn't break the symmetric, revtree-only case.