- Add SharedStorageMonitor thread to periodically verify shared storage:
* Writes a temp file to the shared location and validates MD5 from all nodes.
* Skips nodes with unstable recent heartbeats; retries once; defers decision if any node is unreachable.
* Updates a cluster-wide stateful flag (shared_storage_on) only on conclusive checks.
- New CMAPI endpoints:
* PUT /cmapi/{ver}/cluster/check-shared-storage — orchestrates cross-node checks.
* GET /cmapi/{ver}/node/check-shared-file — validates a given file’s MD5 on a node.
* PUT /cmapi/{ver}/node/stateful-config — fast path to distribute stateful config updates.
- Introduce in-memory stateful config (AppStatefulConfig) with versioned flags (term/seq) and shared_storage_on flag:
* Broadcast via helpers.broadcast_stateful_config and enhanced broadcast_new_config.
* Config PUT is now validated with Pydantic models; supports stateful-only updates and set_mode requests.
- Failover behavior:
* NodeMonitor keeps failover inactive when shared_storage_on is false or cluster size < 3.
* Rebalancing DBRoots becomes a no-op when shared storage is OFF (safety guard).
- mcl status improvements: per-node 'state' (online/offline), better timeouts and error reporting.
- Routing/wiring: add dispatcher routes for new endpoints; add ClusterModeEnum.
- Tests: cover shared-storage monitor (unreachable nodes, HB-based skipping), node manipulation with shared storage ON/OFF, and server/config flows.
- Dependencies: add pydantic; minor cleanups and logging.
* feat(cmapi): add read_only param for API add node endpoint
* style(cmapi): fixes for string length and quotes
Add dbroots of other nodes to the read-only node
On every node change adjust dbroots in the read-only nodes
Fix logging (trace level) in tests
Remove ExeMgr from constants
Fix tests
Manually remove read-only node from ReadOnlyNodes on node removal (because nodes are only deactivated)
Review fixes (mostly switching to StrEnum analog before py3.11, also changes in ruff config)
Read-only nodes are now called read replica consistently
Don't write hostname into IP fields of the config like PMSx/IPAddr, pmx_WriteEngineServer/IPAddr
We calculate ReadReplicas by finding PMs without WriteEngineServer
In _replace_localhost, replace local IP addrs with resolved IP addrs and local hostnames -- with the resolved hostnames.
ModuleHostName/ModuleIPAddr is kept intact.
Keep only IPv4 in ActiveNodes/DesiredNodes/InactiveNodes
feat: add mock DNS resolution builder for testing hostname/IP mappings
* Fix _add_node_to_PMS: if node is already in PMS, save it to existing items to not miss it during the reconstruction of the list
* Make tests independent from CWD
Fixed for _add_Module_entries
Fixed node removal and tests
Fixes for node manipulation tests
Tracing requests
Custom log factory adds all trace values as one log record parameter (it will be empty if trace values are empty, like in MainThread where there are no incoming requests)
MAJOR: Some logic inside node remove changed significantly using active
nodes list from Columnstore.xml to broadcast config after remove.
[fix] TransactionManager passsing extra, remove and optional nodes arguments to start_transaction function
[fix] commit and rollback methods of TransactionManager adding nodes argument
[fix] TransactionManager using success_txn_nodes inside
[fix] remove node logic to use Transaction manager
[fix] cluster set api key call using totp on a top level cli call
[add] missed docstrings
[fix] cluster shutdown timeout for next release
[add] cluster api client class
[fix] cli cluster_app using cluster api client
[add] ClusterAction Enum
[add] toggle_cluster_state function to reduce code duplication
