- 10 10月, 2018 1 次提交
-
-
由 Matteo Merli 提交于
-
- 08 10月, 2018 4 次提交
-
-
由 joefk 提交于
-
由 Boyang Jerry Peng 提交于
### Motivation Fixes #2743 NPE occurs when there are assignments for workers not currently in the membership. This could happen if the worker crashed before failure detector unassigned the functions assign to it and the scheduler is triggered by other means. ### Modifications Filter out the assignments for worker not in membership.
-
由 Sijie Guo 提交于
*Motivation* base64 is non compatible between python2 and 3 *Changes* Provides a python 2/3 compatible method
-
由 Sijie Guo 提交于
*Motivation* We need an upgrade/backward compatibility story for schema enforcement. Especially around: - `Producers cannot connect without a schema to topics with a schema` *Changes* - provide a flag on brokers to enable schema validation (and disabled it by default). this allows a smooth upgrade on brokers, otherwise, it will break all non-java producers on topics with schema immediately when upgrade to the new version
-
- 07 10月, 2018 1 次提交
-
-
由 Sanjeev Kulkarni 提交于
-
- 06 10月, 2018 6 次提交
-
-
由 Matteo Merli 提交于
### Motivation Expose metrics in Prometheus regarding the number of log messages printed by each process. This will add metrics like: ``` # TYPE log4j2_appender_total counter log4j2_appender_total{cluster="standalone",level="debug"} 0.0 log4j2_appender_total{cluster="standalone",level="warn"} 1.0 log4j2_appender_total{cluster="standalone",level="trace"} 0.0 log4j2_appender_total{cluster="standalone",level="error"} 0.0 log4j2_appender_total{cluster="standalone",level="fatal"} 0.0 log4j2_appender_total{cluster="standalone",level="info"} 181.0 ```
-
由 Sanjeev Kulkarni 提交于
* The log4j should be prefixed by pulsarRootDir * User importlib because use of __import__ is discouraged
-
由 Sijie Guo 提交于
*Motivation* Sometime when we run worker service as part of broker, some pods can be hanging on creating producers to assignment topics. It is unknown whether is it a k8s problem or not. But in general, timeout to fail fast to allow k8s to reschedule the pods. *Changes* Add a timeout logic at creating producers.
-
由 Sijie Guo 提交于
*Motivation* Ideally we should run all integration tests on both cluster mode and standalone mode. However the apache ci can't really afford to do so. so we run all the integration tests on cluster mode. We only run basic validation and test new features (e.g. state) on standalone. *Changes* Add PulsarStandalone related test base and suite and a simmple smoke test. This would set the framework for integration tests for state
-
由 Sanjeev Kulkarni 提交于
* Add a config to allow user code dependencies to be installed at runtime. Enable it only for Kubernetes runtime * Fix bug * Fixed indentation issue * Specify the right cmd line * Install dep only in the temp dir * Queue is not in python3. Thus add a try catch block * Fixed unittest
-
由 Sijie Guo 提交于
*Motivation* getFunctionStat call can happen before spanwer starts the runtime *Changes* Add null checks and also change the order of closing sequence in RuntimeSpawner
-
- 05 10月, 2018 12 次提交
-
-
由 Sijie Guo 提交于
*Motivation* People shouldn't be looking into code to understand how to configure a function worker. We should have all the possible settings available in the configuration file. *Changes* reorganize the function worker configuration file and add missing fields
-
由 Sijie Guo 提交于
*Motivation* k8s runtime doesn't support generating function stats at worker service right now. so skip it for now until that feature is added. *Changes* skip function stats for k8s runtime
-
由 Rajan Dhabalia 提交于
### Motivation If pulsar service starts broker and worker in the same process then worker should not use broker-advertised address but use localhost address to connect.
-
由 Sijie Guo 提交于
* [functions][stats] NPE in FunctionStatsGenerator when worker service is not ready *Motivation* NullPointerException was thrown when function worker is running as part of broker and metrics collection kicks in before worker service completes initialization *Changes* Only generate functions when worker service is ready * Fix FunctionSTatsGeneratorTest
-
由 Sanjeev Kulkarni 提交于
* Download the user code into pulsarRootDir * Apply it everywhere * Fix unittest
-
由 Matteo Merli 提交于
-
由 Sanjeev Kulkarni 提交于
-
由 Matteo Merli 提交于
* Removed shading relocations for Circe-checksum and lz4 libraries * Also remove the jni shading in pulsar-broker-shaded * Removed remaining relocations directives * Moved Kafka wrapper tests to integration tests * Do not shade io.netty.buffer.ByteBuf since it's passed between components * Fixed presto dependencies license file
-
由 Matteo Merli 提交于
-
由 Sijie Guo 提交于
- Tick doesn't implement equals - TestBasicPresto doesn't run
-
由 Yuta Mochizuki 提交于
-
由 Rajan Dhabalia 提交于
-
- 04 10月, 2018 14 次提交
-
-
由 Matteo Merli 提交于
-
由 massakam 提交于
For a topic with metadata similar to https://github.com/apache/pulsar/pull/2673, IllegalArgumentException may occur in the following line: https://github.com/apache/pulsar/blob/b2484d92d5068d4f0699eb9c3d31640cb48f9dd0/managed-ledger/src/main/java/org/apache/bookkeeper/mledger/impl/ManagedCursorImpl.java#L655 This is the broker log when the exception has occurred: [invalid_range_error.txt](https://github.com/apache/pulsar/files/2442924/invalid_range_error.txt) It is because `readPosition` is ahead of `ledger.getLastPosition().getNext()`, so `ManagedCursorImpl#getNumberOfEntries()` should return 0 as the number of entries to read in that case. I think this issue and https://github.com/apache/pulsar/pull/2673 are the result of that `ledger.getLastPosition()` is no longer the real last of the managed ledger because of https://github.com/apache/pulsar/pull/1550.
-
由 Sanjeev Kulkarni 提交于
* First cut of the serverside valiation * Revert unneeded changes * Do the easier checks first * Check for null * Simplify * Do the actual validation * Check for null runtime * Update Test * Fix build * Fix build * misc fixes * Add explicit check for serde * Changed for a better condition * Better serde check * Fixed some unittests * Fixed more tests * Fixed unittest * Fixed unittests * Fixed unittest * Stricter checks
-
由 Boyang Jerry Peng 提交于
* adding metrics to presto pulsar connector * rename batch size * adding comments * refactoring metrics * modifying bytes read metric * fixing tests * deleting tmp file * adding jars to LICENSE
-
由 Boyang Jerry Peng 提交于
-
由 Grant Wu 提交于
* Clarify version match requirements between Go and C++ client librarires * Document how to see local changes * Link to docs README * Move warning out from `go get` paragraph
-
由 Matteo Merli 提交于
-
由 Sijie Guo 提交于
*Motivation* Currently trigger function is broken due to the schema enforcement we added recently: A producer without schema can't produce messages into a topic with schema. *Changes* - Rename `AUTO` to `AUTO_CONSUME` - Introduce `AUTO_PRODUCE` schema. The schema produces `byte[]`, but it will validate the bytes are compatible with the schema associated with the topic before producing. - Change trigger function to use `AUTO_PRODUCE` schema.
-
由 Sanjeev Kulkarni 提交于
### Motivation This pr will allow functions being deployed as kubernetes jobs have installation defined labels.
-
由 Rajan Dhabalia 提交于
-
由 Rajan Dhabalia 提交于
-
由 Matteo Merli 提交于
* Do not register each DispatchRateLimiter for policies notifications * Fixed updates on per-subscriptions limits * Fixed testClusterPolicyOverrideConfiguration
-
由 Sijie Guo 提交于
* Fix stats generator * Set the response to text/plain * fix TopicStats and NamespaceStats
-
由 Jia Zhai 提交于
### Motivation add kafka source connect adaptor for debezium This will save data from kafka source connect into Pulsar. ### Modifications add class and test ### Result ut pass
-
- 03 10月, 2018 2 次提交
-
-
由 Sijie Guo 提交于
### Motivation Create consumers before producing will make sure consumers received all the messages that they expect to receive.
-
由 Rajan Dhabalia 提交于
* Avoid scheduling heartbeat function if owner-worker not available * fix: heartbeat function place * add heartbeat check at scheduler-manager * create seaparate list for heatbeat functions
-