GameServerRestartBeforeReadyCrash: Run serially #2791
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
zmerlynn
commented
Nov 4, 2022
|
Build Failed 😱 Build Id: bd31e1cf-d67a-46b4-9c9f-7b677db14698 To get permission to view the Cloud Build view, join the agones-discuss Google Group. |
5af0b52
to
14d961a
Compare
|
Build Failed 😱 Build Id: 573ead3a-2882-4a08-b31e-d0213f0ac596 To get permission to view the Cloud Build view, join the agones-discuss Google Group. |
Narrow the race in googleforgames#2445 by running GameServerRestartBeforeReadyCrash serially. See googleforgames#2445 (comment) for a detailed analysis. Does not fix the issue - this is stopgap until we understand how to fix it.
14d961a
to
730f82d
Compare
|
Build Succeeded 👏 Build Id: 119c7fb8-8b7c-4b48-a493-1511a9941efe The following development artifacts have been built, and will exist for the next 30 days:
A preview of the website (the last 30 builds are retained): To install this version:
|
markmandel
approved these changes
Nov 8, 2022
|
[APPROVALNOTIFIER] This PR is APPROVED This pull-request has been approved by: markmandel, zmerlynn The full list of commands accepted by this bot can be found here. The pull request process is described here
Needs approval from an approver in each of these files:
Approvers can indicate their approval by writing |
chiayi
pushed a commit
to chiayi/agones
that referenced
this pull request
Nov 17, 2022
nodepools and regional clusters Updates to release checklist. (googleforgames#2772) * Updates to release checklist. Adding items that showed up in the recent release that were not written down or required better clarification. * Review updates, and some other small tweaks. Co-authored-by: Robert Bailey <[email protected]> Release 1.27.0 (googleforgames#2776) * Release 1.27.0 * Update FAQ on ExternalDNS (googleforgames#2773) The feature flag it points to have been moved to stable, so the link is not useful any more. Also removed notes on ipv6, since they aren't 100% accurate, as we were discussing in googleforgames#2767. * Updates to release checklist. (googleforgames#2772) * Updates to release checklist. Adding items that showed up in the recent release that were not written down or required better clarification. * Review updates, and some other small tweaks. Co-authored-by: Robert Bailey <[email protected]> * Release-changes * Review comment * Review changes Co-authored-by: Mark Mandel <[email protected]> Co-authored-by: Robert Bailey <[email protected]> Version updates (googleforgames#2778) Players in-game metric for when PlayerTracking is enabled (googleforgames#2765) * Check for DeletionTimestamp of fleet and gameserverset before scaling * Add metric to track player count in gameservers * check PlayerStatus is not nil * Update metrics available in docs * Wrong relref path * typo * Change name for players in game metric to player connected. Add player capacity metric. Hide docs until next agones release. * Duplicate metrics table * add gameserver player tracking metrics to fleetViews Co-authored-by: Mark Mandel <[email protected]> Remove generation for swagger Go code and Add static swagger codes for test (googleforgames#2757) Co-authored-by: Mark Mandel <[email protected]> Updated allocation yaml files under examples/ to use selectors Show how to set graceful termination in a game server that is safe to (googleforgames#2780) evict. Avoid retry from allocateFromLocalCluster under context kill. (googleforgames#2783) * Version updates * issue-2736-changes Co-authored-by: Mark Mandel <[email protected]> Bring SDK base image to debian:bullseye (googleforgames#2769) * Bring SDK base image to debian:bullseye The upgrade to gRPC solved one issue, and I also added a limit to number of processes that could run for `make -j` otherwise the whole thing would fall over (also would crash my dev machine!). Closes googleforgames#2224 * Force refresh of cpp cache on Cloud Build. * Fixes for CI: * Revert CI cache increment (don't think we need it) * Add shell to cpp image for debugging. * Fix formatting issue that is breaking CI. Co-authored-by: Robert Bailey <[email protected]> Update health-checking.md (googleforgames#2785) Fixed spell error: spec.health.failureTheshold to spec.health.failureThreshold Updated allocation yaml files under examples/ to use selectors (googleforgames#2787) Cleanup of load tests (googleforgames#2784) * issue-2744 updated changes with new description * 2744 review changes Sync Pod host ports back to GameServer in GCP (googleforgames#2782) This is the start of the implementation for googleforgames#2777: * Most of this is mechanical and implements a thin cloud product abstraction layer in pkg/cloud, instantiated with New(product). The product abstraction provides a single function so far: SyncPodPortsToGameServer. * SyncPodPortsToGameServer is inserted as a hook while syncing IP/ports, to let different cloud providers handle port allocation slightly differently (in this case, GKE Autopilot) * In GKE Autopilot, we look for a JSON string like `{"min":7000,"max":8000,"portsAssigned":{"7001":7737,"7002":7738}}` as an indication that the host ports were reassigned (per policy). As a side note to anyone watching, this is currently an unreleased feature. If we see this, we use the provided mapping to map the host ports in the GameServer.Spec. With this change, it's possible to launch a GameServer and get a healthy GameServer Pod by adding the following annotation: ``` annotations: cluster-autoscaler.kubernetes.io/safe-to-evict: "true" autopilot.gke.io/host-port-assignment: '{"min": 7000, "max": 8000}' ``` If this PR causes any issues, the cloud product auto detection can be disabled by setting `agones.cloudProduct=generic`, or forced to GKE Autopilot using `agones.cloudProduct=gke-autopilot`. In a future PR, I will add the host-port-assignment annotation automatically on Autopilot Co-authored-by: Mark Mandel <[email protected]> Update gke terraform files to allow autoscaling Fix (not really) problems reported by VSCode (googleforgames#2790) VSCode reports `main redeclared` between allocationload.go and runscenario.go due to the fact that they both look like `package main` binaries in the same directory, similar e.g. [this poster on a different project](https://stackoverflow.com/questions/66970531/vs-code-go-main-redeclared-in-this-block) To fix it, it's easy enough to just give these binaries their own package path and fix up the calling scripts. Along the way, fix a lint complaint in runscenario.go Add location variable for cluster location argument Minor fix changed default of location var to empty string GameServerRestartBeforeReadyCrash: Run serially (googleforgames#2791) Narrow the race in googleforgames#2445 by running GameServerRestartBeforeReadyCrash serially. See googleforgames#2445 (comment) for a detailed analysis. Does not fix the issue - this is stopgap until we understand how to fix it. Enable fieldalignment linter, then mostly ignore it (googleforgames#2795) Enable the fieldalignment linter by enabling all `govet` checks except shadowing. Ignore large swaths of code (tests, cmd/, APIs), and nolint'd existing complaints that seemed irrelevant. Along the way: * removed existing nolint:maligned, as `maligned` is no more. * disabled `structcheck` and `deadcode` as they are deprecated (and I think have been subsumed by other linters?) * changed `gameServerCacheEntry` to `gameServerCache`. It is the cache, not just an entry. * fixed alignment of `gameServerSetCacheEntry`. Add fswatch library to watch and batch filesystem events, use in allocator (googleforgames#2792) This pull refactors the fsnotify code in allocator/main out to a shared library, and in that shared library implements a batched notification processor. Closes googleforgames#1816: This takes a slightly different approach than specified in the issue, instead choosing to just delay processing until after a batch processing period. I chose 1s - it's far longer than necessary, but still much shorter than it takes for the secret changes to propagate to the container anyways. I considered the approach in googleforgames#1816 of trying to parse the actual events, but it's too fiddly to get exactly right: e.g. maybe you only refresh on "write", but then "chmod" could make the file readable whereas it wasn't before, "rename" could expose a file that wasn't there before, etc. Cloud product: Split port allocators, implement Autopilot port allocation/policies (googleforgames#2789) In the Agones on GKE Autopilot implementation, we have no need for the port allocator - the informer/etc. is an unnecessary moving piece. This PR allows for cloud products to provide their own port allocation implementation, and implements the GKE Autopilot "allocator". We do this by: * Splitting portallocator off to its own package. It was basically self-sufficient anyways, except it was a little too friendly with controller_test.go. I solved that by introducing a TestInterface for controller_test.go to upcast to. * Allow cloud product implementations to define their own port allocator. * Defining a new port allocator for GKE that does a simple per-port HostPort allocation, and adds the host-port-assignment annotation to the pod template. * Extend cloudproduct again to add a GameServer validator * And in Autopilot, reject if the PortPolicy is not `Dynamic` Release: Note to switch away from `agones-images` (googleforgames#2809) Since we have few guardrails on accidentally touching `agones-images` project, adding a note in the release checklist to switch back to a local development project after running a release. Flake: TestControllerGameServerCount (googleforgames#2805) Made it deterministic in the test, and got rod of the potential race conditions. Also fix it such that the util function for generating GameServer names always produce a unique name. Closes googleforgames#2804 Co-authored-by: Robert Bailey <[email protected]> Remove Windows FAQ Entry (googleforgames#2811) The contents are no longer accurate, and are covered in the installation section now. Makefile changes for adding location variable added autoscale parameters to Makefile and README Markdown fix in readme Changed LOCATION to always be set with ZONE as default use only if the variable has a value fixed extraneous characters update gke terraform exmaple module Update Node.js dependencies and package (googleforgames#2815) * Update all dependencies and Node,js to LTS version * Update other docker images that use Node.js Added autoscale to example cluster and added to website docs Added defaults and feature expiry Remove zone from gke/variable.tf file.
5 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What type of PR is this?
/kind bug
What this PR does / Why we need it:
Narrow the race in #2445 by running GameServerRestartBeforeReadyCrash serially. See #2445 (comment) for a detailed analysis.
Does not fix the issue - this is stopgap until we understand how to fix it.