CreateJob, CreateService, and StartJobUpdate now include a rudimentary verification function to check if the call made it to the Aurora Scheduler when the client experiences a timeout.
* Simplifying documentation for getting started: Removed outdated information about install Golang on different platforms and instead included a link to the official Golang website which has more up to date information. Instructions for installing docker-compose have also been added.
* Added documentation to all exported functions and structs.
* Unexported some structures and functions that were needlessly exported.
* Adding golang CI default configuration which can be useful while developing and may be turned on later in the CI.
* Moving build process in CI to xenial.
* Reducing line size. in some files and shadowing in some test cases.
* Adding Aurora URL validator in order to handle scenarios where incomplete information is passed to the client. The client will do its best to guess the missing information such as protocol and port.
* Upgraded to testify 1.3.0.
* Added configuration to fail on a non-temporary error. This is reverting to the original behavior of the retry mechanism. However, this allows the user to opt to fail in a non-temporary error.
* Bumped up version to 1.21.1
* Moving admin functions to a new file. They are still part of the same pointer receiver type.
* Removing dead code and fixing some comments to add space between backslash and comment.
* Adding set up and tear down to run tests script. It sets up a pod, runs all tests, and then tears down the pod.
* Added `--rm` to run tests Mac script.
* Removing cookie jar from transport layer as it's not needed.
* Changing all error messages to start with a lower case letter. Changing some messages around to be more descriptive.
* Adding an argument to allow the retry mechanism to stop if a timeout has been encountered. This is useful for mutating API calls. Only StartUpdate and CreateService have enabled by default stop at timeout.
* Added 2 tests for when a call goes through despite the client timing out. One is with a good payload, one is with a bad payload.
* Updating changelog with information about the error type returned.
* Adding test for duplicate metadata.
* Refactored JobUpdateStatus monitor to use a new monitor called JobUpdateQuery. Update monitor will now still continue if it does not find an update to monitor. Furthermore, it has been optimized to reduce returning payloads from the scheduler as much as possible. This is through using the GetJobUpdateSummaries API instead of JobUpdateDetails and by including a the statuses we're searching for as part of the query.
* Added documentation as to how to handle a timeout on an API request.
* Optimized GetInstancesIds to create a copy of the JobKey being passed down in order to avoid unexpected behavior. Instead of setting every variable name separately, now a JobKey array is being created.
* Ported all code from Thrift 0.9.3 to Thrift 0.12.0 while backporting some fixes from gorealis v2
* Removing git.apache.org dependency from Vendor folder as this dependency has migrated to github.
* Adding github.com thrift dependency back but now it points to github.com
* Removing unnecessary files from Thrift Vendor folder and adding them to .gitignore.
* Updating dep dependencies to include Thrift 0.12.0 from github.com
* Adding changelog.
* End to end tests: Adding coverage for killinstances.
* End to end tests: Deleting instances after partition policy recovers them.
* End to end tests: Adding more coverage to the realis API.
* End to end tests: Allowing arguments to be passed to runTestMac so that '-run <test name>' can be passed in.
* End to end tests: Reducing the resources used by CreateJob test.
* End to end tests: Adding coverage for Pause and Resume update.
* End to end tests: Removed checks for Aurora_OK response as that should always be handled by the error returned by the API. Changed names to be less verbose and repetitive.
* End to end tests: Reducing watch time for instance running when creating service for reducing time it takes to run end to end test.
* Add trace level to print out response thrift objects. Allows user to control whether these are printed or not to avoid pollution.
* Using named parameters to be more explicit about what is being set for LevelLogger.
* Adding TracePrint and TracePrintln. Inlined library level prefixes.
* Detecting if the transport error was not temporary in which case we stop retrying. Changed bug where get results was being called before we checked for an error.
* Adding exception for EOF error. All EOF errors will be retried.
* Addressing race conditions that may happen when client is closed or connection is re-established.
* Adding documentation about how this particular implemantion of the realis client uses retries in scenarios where a temporary error is found.
* Bugfix: switch statements were missing fallthrough statement thus making them retry non-retriable errors. Using a list to catch cases now.
* Adding tests for CreateService, createService when the executor doesn't exist, and createJob when the executor doesn't exist. Renamed Pulse test to reflect that it's using CreateService instead of CreateJob.
* Repsonse propagate back up to caller for context for CreateJob, CreateService, and StartJobUpdate.
* Deleting PR template as Travis CI takes care of running tests and formatting tests now.
* Fixing possible race condition when passing backoff around as a pointer.
* Adding a debug logger that is turned off by default. If debug is turned on, but a logger has not been assigned, a default logger that will print to STDOUT will be created.
* Making Mutex a pointer so that there's no chance it can accidentally be copied.
* Removing a leftover helper function from before we changed how we configured the client.
* Minor changes to demonstrate how a logger can be used in conjunction to debug mode in the sample client.
* ZK retries have been cleaned up. We will now retry after every error
EXCEPT when we have a badly formed path.
* ZK library has been reworked with optional arguments pattern to not be
so intertwined with the cluster.json file.
* Timeout error has been re-implemented as RetryError. RetryError
behaves like a Timeout error but is used exclusively to add more context
privately. This allows us to have unit tests that check our retry
mechanism is actually retrying.
* Additional logging has been added to retry mechanisms as well as to
the Zookeeper library we use.
* Deleting permament error as it doesn't make sense. Just return a plain old error and that will be considered permanent.
* Removing double closure at as it's unmaintainable and can be error prone. Separated back offs into a generic one and a thrift call specific one.
* ZK leader finder now returns a temporary error instead of constantly no leader found and quitting. It could be that the leader info is being propagated so it's worth trying another time.
* Adding more logging to the retry.
* Wrapping lock and unlock in an anonymous function so that we can use defer on unlock such that it is called in the case of a panic.
* Fixing logic that can lead to nil error being returned and retry stopping early.
* Fixing possible code path that may lead to an incorrect nil error.
* Changing incorrect license in some source files.
* Changing CreateService to mimic CreateJob by setting the batch size to the instance count.
* Changing Getcerts to GetCerts to match the style of the rest of the codebase.
* Overhauled error handling. Backoff now recognizes temporary errors and continues to retry if it finds one.
* Changed thrift function call wrapper to be more explicitly named and to perform more safety checks.
* Moved Jitter function from realis to retry.
* API code is now more uniform and follows a certain template.
* Lock added whenever a thrift call is made or when a modification is done to the connection. Note that calling ReestablishConn externally may result in some race conditions. We will move to make this function private in the near future.
* Added test for Realis session thread safety. Tested ScheduleStatus monitor. Tested monitor timing out.
* Returning nil whenever there is an error return so that there are no ambiguities.
* Using defer with unlock so that the lock is still released if a panic is invoked.
