-
Notifications
You must be signed in to change notification settings - Fork 1.1k
[hitless upgrade] Support for none moving-endpoint-type in maintenance event notifications (CAE-1285) #3396
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[hitless upgrade] Support for none moving-endpoint-type in maintenance event notifications (CAE-1285) #3396
Conversation
none indicates that the MOVING message doesn’t need to contain an endpoint. In such a case, the client is expected to schedule a graceful reconnect to its currently configured endpoint after half of the grace period that was communicated by the server is over.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Pull Request Overview
This PR extends Lettuce's Redis maintenance event notifications to support the moving-endpoint-type=none
option. When this option is used, the MOVING message doesn't contain an endpoint and triggers a graceful reconnect to the currently configured endpoint after half the grace period has elapsed.
- Added
NONE
enum value toAddressType
with proper documentation - Implemented deferred rebind logic that schedules reconnection after half the grace period
- Modified MOVING event parsing to handle null endpoints correctly
Reviewed Changes
Copilot reviewed 4 out of 4 changed files in this pull request and generated 3 comments.
File | Description |
---|---|
MaintenanceEventsOptions.java |
Added NONE address type enum value with documentation |
RedisHandshake.java |
Added mapping for NONE address type to "none" string |
MaintenanceAwareConnectionWatchdog.java |
Implemented deferred rebind logic and null endpoint handling |
MaintenanceAwareConnectionWatchdogUnitTests.java |
Added test coverage for null endpoint scenario and refactored helper methods |
Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.
src/main/java/io/lettuce/core/protocol/MaintenanceAwareConnectionWatchdog.java
Outdated
Show resolved
Hide resolved
src/main/java/io/lettuce/core/protocol/MaintenanceAwareConnectionWatchdog.java
Outdated
Show resolved
Hide resolved
src/test/java/io/lettuce/core/protocol/MaintenanceAwareConnectionWatchdogUnitTests.java
Outdated
Show resolved
Hide resolved
Co-authored-by: Copilot <[email protected]>
null for rebind adress can be returned as part of MOVING notification if client is connected using 'moving-endpoint-type=none'
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks good, thanks!
…e event notifications (CAE-1285) (#3396) * support MOVING with none none indicates that the MOVING message doesn’t need to contain an endpoint. In such a case, the client is expected to schedule a graceful reconnect to its currently configured endpoint after half of the grace period that was communicated by the server is over. * formatting * Apply suggestions from code review Co-authored-by: Copilot <[email protected]> * Fix NPE * Add test to cover null rebindAddress null for rebind adress can be returned as part of MOVING notification if client is connected using 'moving-endpoint-type=none' * Add java docs to RebindAwareAddressSupplier --------- Co-authored-by: Copilot <[email protected]>
* [Hitless Upgrades] React to maintenance events (#3345) * v0.1 * Simple reconnect now working * Bind address from message is now considered * Self-register the handler * Format code * Filter push messages in a more stable way * (very hacky) Relax comand expire timers globbaly * Configure if timeout relaxing should be applied * Proper way to close channel * Configure the timneout relaxing * Sequential handover implemented * Did not address formatting * Prolong the rebind windwow for relaxed tiemouts * PubSub no longer required; CommandExpiryWriter is now channel aware; Polishing * Use the new MOVING push message from the RE server * Unit test was not chaining delgates in the same way that the RedisClient/RedisClusterClient was * Fix REBIND message validation * Fixed the expiry mechanism * Polishing * Fix NPE. Seems like AttributeMap.attr is not accurate and actually return's null causing some unit test failures. * Add support for MIGRATING/MIGRATED message handling in command expiry This commit adds the ability to listen for MIGRATING and MIGRATED messages and trigger extended command expiry timeouts during Redis shard migration. Key changes: - Enhanced RebindAwareConnectionWatchdog to detect MIGRATING/MIGRATED messages - RebindAwareExpiryWriter to trigger timeout relaxation whenever MIGRATING message is received This feature allows commands to have relaxed timeouts during shard migration operations, preventing unnecessary timeouts when Redis is temporarily busy with migration tasks. * formating * Fix Disabling relaxTimeouts after upgrade can interfere with an ongoing one from re-bind * Additional fix for timeout relaxing disabled * Fix push message listener registered multiple times after rebind. * Fix: Report correct command timeout when relaxTimeout is configured * Disable relaxedTimeout after configured grace period - Introduce a delay before disabling relaxedTimeout - Grace period duration is provided via push notification * Code clean up - Remove reading from pub/sub chanel and relay only on push notifications * Add FAILING_OVER/FAILED_OVER * Polishing : Rename components to use the word 'maintenace' --------- Co-authored-by: Igor Malinovskiy <[email protected]> Co-authored-by: ggivo <[email protected]> # Conflicts: # src/main/java/io/lettuce/core/ClientOptions.java * [Hitless upgrades] Add unit tests for the newly introduced classes #3355 (#3356) * Unit tests for the maintanence aware classes * Did not format properly * Proper license * Cae 633 add functional tests notifications (#3357) - excluding JsonExample.java * Cae 1130 timeout tests (#3377) * initial WIP, with lots of debugging, and some non-working tests * debug * more attemtps at debugging * Refactor: Move cluster state management methods from MaintenanceNotificationTest to RedisEnterpriseConfig - Moved refreshClusterConfig, recordOriginalClusterState, and restoreOriginalClusterState methods - Updated call sites in MaintenanceNotificationTest to use RedisEnterpriseConfig methods - Added required imports and static variables to RedisEnterpriseConfig - Maintained existing functionality while improving code organization Improvements to RelaxedTimeoutConfigurationTest: - Simplified traffic generation logic by removing complex multi-phase testing - Streamlined BLPOP command execution with better timeout detection - Added relaxed timeout detection and recording during maintenance events - Improved logging and error handling for timeout analysis - Enhanced test assertions to focus on relaxed timeout detection rather than success counts - Added MOVING operation duration tracking for better test analysis * Improve test reliability and cleanup: Add @AfterEach cleanup, enhance endpoint tracking, and improve logging * add un-relaxed tests. will investigate further why they got broken at some point via diff * CAE-1130: Update timeout configuration test and watchdog implementation * Reset MaintenanceAwareConnectionWatchdog.java and log4j2-test.xml to upstream versions * Clean up debug info and outdated comments from timeout tests - Remove large debug block with reflection-based debugging in RelaxedTimeoutConfigurationTest - Simplify excessive debug logging and verbose markers (*** and ===) - Clean up maintenance notification test logging - Improve push notification monitor message formatting - Maintain all test functionality while improving code readability * Refactor: Move inline comments above code and fix string comparisons - Move all inline comments to be above the code they reference in: * RelaxedTimeoutConfigurationTest.java * RedisEnterpriseConfig.java * MaintenanceNotificationTest.java - Replace string != comparisons with .equals() for proper string comparison - Apply code formatting via Maven formatter This improves code readability and follows Java best practices for string comparison. * [Hitless Upgrades] Zero-server-side configuration with client-side opt-in (#3380) * Support for Client-side opt-in A client can tell the server if it wants to receive maintenance push notifications via the following command: CLIENT MAINT_NOTIFICATIONS <ON | OFF> [parameter value parameter value ...] * update maintenance events to latest format - MIGRATING <seq_number> <time> <shard_id-s>: A shard migration is going to start within <time> seconds. - MIGRATED <seq_number> <shard_id-s>: A shard migration ended. - FAILING_OVER <seq_number> <time> <shard_id-s>: A shard failover of a healthy shard started. - FAILED_OVER <seq_number> <shard_id-s>: A shard failover of a healthy shard ended. - MOVING <seq_number> <time> <endpoint>: A specific endpoint is going to move to another node within <time> seconds * clean up * Update FAILED_OVER & MIGRATED to include additional time field * update is private reserver check & add unit tests update is private reserver check * add unit tests for handshake with enabled maintenance events * add missing copyrights/docs * format * address review comments * Revert address after rebind operation expires * Update event's validation spec - MIGRATING <seq_number> <time> <shard_id-s>: A shard migration is going to start within <time> seconds. - MIGRATED <seq_number> <shard_id-s>: A shard migration ended. - FAILING_OVER <seq_number> <time> <shard_id-s>: A shard failover of a healthy shard started. - FAILED_OVER <seq_number> <shard_id-s>: A shard failover of a healthy shard ended. - MOVING <seq_number> <time> <endpoint>: A specific endpoint is going to move to another node within <time> seconds * rebase * format after rebase * Apply suggestions from code review Co-authored-by: Tihomir Krasimirov Mateev <[email protected]> * javadoc updated * Update src/main/java/io/lettuce/core/internal/NetUtils.java Co-authored-by: Tihomir Krasimirov Mateev <[email protected]> --------- Co-authored-by: Tihomir Krasimirov Mateev <[email protected]> * [hitless upgrade] Support for none moving-endpoint-type in maintenance event notifications (CAE-1285) (#3396) * support MOVING with none none indicates that the MOVING message doesn’t need to contain an endpoint. In such a case, the client is expected to schedule a graceful reconnect to its currently configured endpoint after half of the grace period that was communicated by the server is over. * formatting * Apply suggestions from code review Co-authored-by: Copilot <[email protected]> * Fix NPE * Add test to cover null rebindAddress null for rebind adress can be returned as part of MOVING notification if client is connected using 'moving-endpoint-type=none' * Add java docs to RebindAwareAddressSupplier --------- Co-authored-by: Copilot <[email protected]> * [Hitless Upgrades] Enable maintenance events support by default (CAE-1303) (#3415) * set default relaxed timeout to 10s * Enable maintenance events support by default * Enable maintenance events support by default * fix tests - ensure MaintenanceAwareExpiryWriter is registered for events when wathcdog is created - command timeout was not applied * fix tests - sporadic failure because of timeout of 50ms RedisHandshakeUnitTests.handshakeDelayedCredentialProvider:153 » ConditionTimeout - new command introduced during handshake, increase the timeout to 100ms * resolve errors after rebase on main - reset() - removed with issue#3328 - remove deprecated code from issue#907 (#3395) * Remove LettuceMaintenanceEventsDemo.java - no longer needed. --------- Co-authored-by: Igor Malinovskiy <[email protected]> Co-authored-by: ggivo <[email protected]> Co-authored-by: kiryazovi-redis <[email protected]> Co-authored-by: Copilot <[email protected]>
Description:
This change extends Lettuce’s support for Redis maintenance event notifications by adding the none option for
moving-endpoint-type
when enabling maintenance events.According to the specification,
moving-endpoint-type
determines whether theMOVING
push message contains an internal/external IP or FQDN.The new special value none indicates that the
MOVING
message should not contain any endpoint; instead, the endpoint field is null.When the client receives a MOVING message with moving-endpoint-type=none, it will schedule a graceful reconnect to the currently configured endpoint after half of the grace period specified in the message has elapsed.
Summary of changes:
Example to enable maintenance notifications with none
CLIENT MAINT_NOTIFICATIONS ON moving-endpoint-type none
Tested manually & unit test
Make sure that:
mvn formatter:format
target. Don’t submit any formatting related changes.