- 30 3月, 2019 40 次提交
-
-
由 Jia Zhai 提交于
-
由 Rajan Dhabalia 提交于
-
由 Sijie Guo 提交于
*Motivation* The script doesn't update the parent version for protobuf-shaded module *Modifications* use `versions:update-parent` to update parent version
-
由 Sijie Guo 提交于
*Motivation* Topics count is a namespace level metric. Currently if `exposeTopicLevelMetricsInPrometheus=true` we don't expose `pulsar_topics_count`. However `pulsar_topics_count` is a very useful metric for monitoring the healthy of the cluster. *Modifications* if `exposeTopicLevelMetricsInPrometheus=true`, we expose namespace level metrics that are not exposed at topic level. e.g. `pulsar_topics_count`.
-
由 Sijie Guo 提交于
*Motivation* Currently integration tests are broken due to #3827 `PULSAR_` is used almost everywhere in our scripts. If we are using `PULSAR_` for appending new keys, then we are potentially appending a wrong key into the config file. Currently this happens on presto. We are appending `PULSAR_ROOT_LOGGER` into presto's config file and cause presto fail to load the config. *Modifications* Use `PULSAR_PREFIX_` instead of `PULSAR_`
-
由 Sanjeev Kulkarni 提交于
* Add config variables if absent * Took feedback into account
-
由 Matteo Merli 提交于
-
由 Ezequiel Lovelle 提交于
Same fix as #3746 but for cpp client. - Filter consumers for the same topic name. - Add test to verify that same subscription names with different topics are allowed to be different consumers subscription instead of reused.
-
由 Tevic 提交于
* fix message_id_serialize to empty slice * test for serialize and deserialize
-
由 Kai 提交于
-
由 Matteo Merli 提交于
* In Java allow messages that compress to <5mb to be sent with batching enabled * Test is not anymore accurate since behavior changed and we're already testing in ProducerConsumerTest
-
由 Sijie Guo 提交于
*Motivation* Fixes #3803 Hardcoding is a very bad practice. It means we have no way to alter system behavior when production issues occur. *Modifications* introduce a few read batch related settings to make them configurable
-
由 Ezequiel Lovelle 提交于
* Feature - implement reference count for ConsumerImpl Add reference count for ConsumerImpl in order to track reused instances of a consumer instance returned by `subscribe()` method call. Having the reference of subscribed consumer instances offers the ability to not close a consumer until the last corresponding `close()` is being called. Modifications: - Add field on ConsumerBase to track references of consumer instances subscribed by the user. - Add checks on ConsumerImpl to know whether close() action should be performed regarding of reference count being zero value. - Increment reference count when a previous built consumer instance is being used by caller. Future considerations: When optimization #3312 is going to be made for other consumers implementation aside from ConsumerImpl it should add refCount checks on close() method call. * Add tests for reference count on ConsumerImpl - Add test to verify ConsumerImpl reference count on close() method. - Fix test from dup consumers feature with refcount.
-
由 Rajan Dhabalia 提交于
* [pulsar-webscoket] avoid creating temp list of metrics on every-metrics generator * fix non-sync getMetrics
-
由 Sijie Guo 提交于
*Motivation* Fixes #3686
-
由 Sijie Guo 提交于
Fixes #3734 *Motivation* Exception occurred when using `BytesSchema.of()` ``` Exception in thread "main" java.lang.ExceptionInInitializerError at org.apache.pulsar.examples.simple.ProducerExample.main(ProducerExample.java:32) Caused by: java.lang.RuntimeException: java.lang.reflect.InvocationTargetException at org.apache.pulsar.client.internal.ReflectionUtils.catchExceptions(ReflectionUtils.java:36) at org.apache.pulsar.client.internal.DefaultImplementation.newKeyValueSchema(DefaultImplementation.java:158) at org.apache.pulsar.client.api.Schema.<clinit>(Schema.java:123) ... 1 more Caused by: java.lang.reflect.InvocationTargetException at sun.reflect.NativeConstructorAccessorImpl.newInstance0(Native Method) at sun.reflect.NativeConstructorAccessorImpl.newInstance(NativeConstructorAccessorImpl.java:62) at sun.reflect.DelegatingConstructorAccessorImpl.newInstance(DelegatingConstructorAccessorImpl.java:45) at java.lang.reflect.Constructor.newInstance(Constructor.java:423) at org.apache.pulsar.client.internal.DefaultImplementation.lambda$newKeyValueSchema$16(DefaultImplementation.java:160) at org.apache.pulsar.client.internal.ReflectionUtils.catchExceptions(ReflectionUtils.java:34) ... 3 more Caused by: java.lang.NullPointerException at org.apache.pulsar.client.impl.schema.KeyValueSchema.<init>(KeyValueSchema.java:68) ... 9 more ``` The problem introduced because the weird class loading and reflection sequence. When accessing `BytesSchema`, `BytesSchema` will try to initialize `Schema`. When initializing Schema, it will attempts to initialize `KV_BYTES` using reflection, and initializing KV_BYTES requires `BytesSchema`. Hence it causes KV_BYTES not being initialized correctly. The change is to avoid this recursive class loading.
-
由 Matteo Merli 提交于
### Motivation Fixes #3779 The behavior of acknowledgment has always been to throw exception if consumer is not connected. In reality it doesn't make much sense to bubble back exception to user in this case because: * There's no ack on the ack anyway * The only reason for ack to fail is if consumer is disconnected * If consumer is disconnected, messages are going to be replayed in any case * Therefore, there's no action to do when acking while disconnected This is already the case in the Java and Go clients (additionally Java client performs client side grouping of acks based on time). Changing the Python client behavior to be the same.
-
由 Matteo Merli 提交于
-
由 Matteo Merli 提交于
-
由 Matteo Merli 提交于
-
由 Boyang Jerry Peng 提交于
* fix: function config cleanupSubscription update bug * add and fix unit tests
-
由 Boyang Jerry Peng 提交于
* propagate default resource values in Pulsar Functions * fix unit tests
-
由 Matteo Merli 提交于
* Fixed Reader.HasNext() in Go client * Fixed formatting * Removed commented code
-
由 Ezequiel Lovelle 提交于
Fixes #3743 issue. Return previous instance of a consumer in the subscription processed should only be considered with the scope of the same topic. Modifications: - Remove optimization of duplicated consumers for multi topics subscribe and pattern topics subscribe, this should be handled with a different approach. - Filter consumers for the same topic name. - Filter consumers which are connected to broker, this is not necessary to fix this issue but is a good thing to do. - Add test that verifies that same subscription will allow different consumers instance for different topics.
-
由 Sanjeev Kulkarni 提交于
* Expose Secrets in Source/SinkContext * Fix unittest
-
由 Sanjeev Kulkarni 提交于
* Allow users to update everything in inputspecs except for isregexpattern * Added more tests and fixed a bug caught by them
-
由 Matteo Merli 提交于
### Motivation Use at least 8 threads to avoid having Jetty go into threads starving and having the possibility of getting into a deadlock where a Jetty thread is waiting for another HTTP call to complete in same thread. This solve the issues of requests timing out when the broker is making REST calls to itself. Such examples are when running in standalone mode and creating a function.
-
由 Sanjeev Kulkarni 提交于
* Retry creation of assignment topic a few times before giving up * Use Action based retry mechanism * Fix build * Catch interrupted exception * Fix unittest * Added header
-
由 Matteo Merli 提交于
-
由 Boyang Jerry Peng 提交于
* fix function termination cleanup * cleaning up
-
由 Sanjeev Kulkarni 提交于
* Simplified the workflow of functionruntime manager * Fix unittest * Took feedback into account * added missing imports
-
由 Sanjeev Kulkarni 提交于
* On publish failures, log error and count them as sys exceptions * Took feedback
-
由 Sanjeev Kulkarni 提交于
-
由 Sanjeev Kulkarni 提交于
-
由 nkurihar 提交于
-
由 Matteo Merli 提交于
* When the loadmanager leader is not available, fall through regular least loaded selection * Handle exceptions coming from mock zk in tests
-
由 Rajan Dhabalia 提交于
-
由 Rajan Dhabalia 提交于
-
由 Sanjeev Kulkarni 提交于
-
由 Boyang Jerry Peng 提交于
-