[tplavcic]

This covers scenarios from PS-11378 ticket:

• (1) Provision new node by restoring backup taken using Xtrabackup from an online node and join the GR node into the cluster.
• (6) Disconnect network on one of secondary nodes and later reconnect it within gr_member_expel_timeout and group_replication_unreachable_majority_timeout seconds and verify that the node is not expelled and it will be back online
• (7) Disconnect network on primary node and later reconnect it within gr_member_expel_timeout and group_replication_unreachable_majority_timeout seconds and see that the node is not expelled and it will be back online
• (8) Disconnect network on one of secondary nodes and later reconnect it within gr_member_expel_timeout but after group_replication_unreachable_majority_timeout seconds and see that member expels and it automatically retries joining and successfully joins back as per group_replication_autorejoin_tries(default 3).
• (9) Disconnect network on primary node and later reconnect it within gr_member_expel_timeout but after group_replication_unreachable_majority_timeout seconds and see that member expels and it automatically retries joining and successfully joins back as per group_replication_autorejoin_tries(default 3). See that a new primary is elected successfully.
• (10) Disconnect network on primary node and later reconnect it after gr_member_expel_timeout but within group_replication_unreachable_majority_timeout seconds and see that member expels and it automatically retries joining and successfully joins back as per group_replication_autorejoin_tries(default 3). See that a new primary is elected successfully.
• Test for asynchronous connection between the two GR clusters in case of primary failover in source and replica cluster. (Async channel running with SOURCE_CONNECTION_AUTO_FAILOVER = 1)

[Copilot]

Copilot review overview

🟡 Changes recommended

Async metadata replication and inherited clone history can cause the new tests to fail incorrectly.

Review effort: Balanced
Findings: 1 · 2

Open (3)

• Exclude local cluster metadata from replica channel · New
• Collect failure logs from gr_cluster_pair fixtures · New
• Snapshot clone status before asserting incremental recovery · New

What changed in this PR

Adds PS-11378 coverage for backup-based provisioning, Group Replication isolation timeouts, autorejoin behavior, and asynchronous cluster failover.

Changes:

• Adds XtraBackup provisioning and timeout-isolation scenarios.
• Adds asynchronous replication failover tests between two GR clusters.
• Extends GR helpers for restored nodes, server-ID offsets, and async channels.

File	Description
xtrabackup_helper.py	Supports preparing full-only backups.
test_provision_from_backup.py	Tests provisioning and joining a restored node.
test_isolation_timeouts.py	Covers timeout and autorejoin scenarios 6–10.
test_async_failover.py	Tests source and replica primary failover.
group_replication_helper.py	Adds provisioning, timeout, and async replication helpers.
conftest.py	Adds a two-cluster fixture.

---

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

[Copilot]

Copilot review overview

🟡 Changes recommended

Replica-side async failover is not enabled, and restarted replica members can apply unfiltered metadata before filter restoration.

Review effort: Balanced
Findings: 1

Open (1)

• Restore replication filter before allowing replica threads to start · New

Resolved since last review (3)

• Exclude local cluster metadata from replica channel
• Snapshot clone status before asserting incremental recovery
• Collect failure logs from gr_cluster_pair fixtures

Previously missed (2)

In code that hasn't changed since last review

Enable automatic channel startup on elected replica primary

test_scripts/​ps/​group-replication/​test_async_failover.py:45

SOURCE_CONNECTION_AUTO_FAILOVER can switch a running receiver to another source, but it does not start this channel when a different replica-cluster member is elected primary. Because START REPLICA is issued only on the current primary, the failover="replica" case will leave the new primary's channel stopped unless group_replication_start_failover_channels is enabled on the replica members. Set it before configuring/starting the channel.

Use effective server ID in startup logs and command

test_scripts/​ps/​group-replication/​group_replication_helper.py:1455

With server_id_offset, this startup command now uses IDs such as 101–103, but the preceding log still reports the unoffset index (1–3). That makes the diagnostics misleading in the new two-cluster tests. Compute the effective server ID once and use it in both the log and command.