User Channel History Compaction tests

[vipbhardwaj]

This PR is the exact same motive as #495, but focuses on User Channel History Compaction.
The tests are also focusing on the same, they are similar to Document CHC, but written from User CHC point of view.
There are a couple of regression tests, but mostly its Unit and Functional testcases which span across CBS-SGW-CBL and verify things from CBL point of view for users as well.
This PR relies on a file pushed in #495, since the backfill was used by the test in that file as well, but after discussion with @RIT3shSapata and @bbrks , that single test was removed, as seen in one recent commit in that PR.
Hence the file is in a different PR and is being used in a different PR.

[Copilot]

Pull request overview

Adds QE coverage and specifications for Sync Gateway user access-history compaction.

Changes:

• Adds tests covering collections, roles, concurrency, and offline-client behavior.
• Documents test scenarios and expected endpoint behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Review summary
tests/QE/test_user_access_history_compaction.py	Critical: configure collection sync functions for channel membership. Moderate: use a continuous replicator or narrow the live-session assertions.
spec/tests/QE/test_user_access_history_compaction.md	Documents the user access-history compaction test scenarios.

Suppressed comments (10)

spec/tests/QE/test_user_access_history_compaction.md:140

• The possessive is punctuated as 'frank's, which makes the quoted username malformed in this step. Rephrase it as “the access history for user 'frank'”.

5. Get user 'frank's access history and check that channel 'ROLE_CHAN' is absent (the

spec/tests/QE/test_user_access_history_compaction.md:199

• The possessive is punctuated as 'iris's, leaving the quoted username malformed. Rephrase the step as “the access history for user 'iris'”.

4. While that session is still open, compact channel 'A' out of user 'iris's access
   history.

spec/tests/QE/test_user_access_history_compaction.md:310

• The possessive is punctuated as 'leo's, leaving both quoted usernames malformed in this step. Rephrase it as “the access of user 'leo'” and “that user's access history.”

5. While offline, revoke user 'leo's access to channel 'A', then compact channel 'A' out
   of user 'leo's access history before the client reconnects.

tests/QE/test_user_access_history_compaction.py:517

• The result of the compaction is only dumped and never asserted, so this test still passes if the endpoint returns an empty/no-op result and the live-session scenario is not actually exercised. Assert that A was reported as compacted (and, ideally, that it is absent from the subsequent access-history GET) before relying on the later B visibility check.

            compacted = await sg.compact_user_access_history(db_name, "iris", {"_default": {"_default": ["A"]}})
            self._dump("compact_user_access_history/iris while session is live", compacted)

tests/QE/test_user_access_history_compaction.py:647

• This test never verifies that the db1 compaction succeeded; it only checks that db2 was unaffected. A broken or no-op compaction would therefore pass because db2 would still contain A either way. Assert that the response reports A as compacted (and/or verify db1 history is empty) before checking isolation.

        self.mark_test_step("Compact channel 'A' for that user on the first database only")
        compacted = await sg.compact_user_access_history("db1", "shared_name", {"_default": {"_default": ["A"]}})
        self._dump("compact_user_access_history/shared_name on db1", compacted)

tests/QE/test_user_access_history_compaction.py:350

• The test never verifies that frank actually inherits ROLE_CHAN before the role is updated. If add_role or the user's admin_roles assignment were ignored, every later assertion (empty user history, empty user compaction, and a missing role endpoint) would still pass. Create a document in ROLE_CHAN and assert it is visible through a client authenticated as frank before removing the role's access.

        self.mark_test_step("Create user 'frank' assigned to role 'myrole', with no direct channel access of their own")
        await sg.add_user(
            db_name,
            "frank",
            password="pass",
            admin_roles=["myrole"],
            collection_access={"_default": {"_default": {"admin_channels": []}}},
        )

tests/QE/test_user_access_history_compaction.py:470

• Any CblSyncGatewayBadResponseError is accepted as the expected concurrent loser, including unrelated 400/401/404 responses or a server error unrelated to the race. That makes this regression pass when the endpoint is malformed or unauthorized. Restrict the accepted status to the conflict-style codes used by the analogous document-compaction test (409/500).

            assert isinstance(failure, CblSyncGatewayBadResponseError)

tests/QE/test_user_access_history_compaction.py:357

• The possessive is punctuated as 'frank's, which makes the quoted username malformed. Rephrase it as “the access history for user 'frank'” (or use frank's without surrounding quotes).

            "Get user 'frank's access history and check that channel 'ROLE_CHAN' is absent "
            "(the history lives on the role's own principal record, not the user's)"

tests/QE/test_user_access_history_compaction.py:514

• The possessive is punctuated as 'iris's, leaving the quoted username malformed. Rephrase this as “the access history for user 'iris'” to make the step unambiguous.

                "While that session is still open, compact channel 'A' out of user 'iris's access history"

tests/QE/test_user_access_history_compaction.py:705

• The possessive is punctuated as 'leo's, leaving the quoted username malformed in this test step. Rephrase it as “the access of user 'leo'” and refer to “that user's” history.

                "While offline, revoke user 'leo's access to channel 'A', then compact channel 'A' out of "
                "the user's access history before the client reconnects"

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[torcolvin]

Every single-collection call site collapses to one line:

Suggested change

	await _setup_db(sg, cblpytest.clusters[0].sync_gateway_cluster, cbs, db_name, "data-bucket")
	await cblpytest.clusters[0].create_database(db_name, _ACCESS_TRACKING_CONFIG)

Same at lines 140, 156, 181, 231, 360, 455, 499, 576, 611 and 699.

The mark_test_step text above stays accurate -- Cluster.create_database does create the bucket.

The two-database test at lines 657-658 needs a different bucket per database; DatabaseConfig is a pydantic model, so:

await cblpytest.clusters[0].create_database("db1", _ACCESS_TRACKING_CONFIG.model_copy(update={"bucket": "data-bucket-1"}))
await cblpytest.clusters[0].create_database("db2", _ACCESS_TRACKING_CONFIG.model_copy(update={"bucket": "data-bucket-2"}))

[torcolvin]

The three tests that need an extra collection pass their config inline, the way #495 handled its one multi-collection test:

Suggested change

	await _setup_db(
	sg, cblpytest.clusters[0].sync_gateway_cluster, cbs, db_name, "data-bucket", extra_collections=["other"]
	)
	await cblpytest.clusters[0].create_database(
	db_name,
	DatabaseConfig(
	bucket=_BUCKET,
	index=IndexConfig(num_replicas=0),
	scopes={
	"_default": ScopeConfig(
	collections={
	"_default": {"sync": _CHANNEL_SYNC_FUNCTION},
	"other": {"sync": _CHANNEL_SYNC_FUNCTION},
	}
	)
	},
	),
	)

Same at lines 309 and 413.

[torcolvin]

Suggested change

	case from a user-history angle). Uses the shared `backfill_after_offline` helper
	(`tests/shared/backfill_after_offline.py`) to simulate the offline window for both the
	case from a user-history angle). Uses `_one_shot_pull` helper to
	simulate the offline window -- the gap between the two one-shot pulls -- for both the

[torcolvin]

Can you pick up the changes from the stacked PR to remove this function?

[torcolvin]

this I think will need wait_for_caching_feed=True. The compaction will work without the cahing feed but wait_for_all_documents will not.

[torcolvin]

#516 will create sg.get_user_client()