Matteo Merli comments

Results 70 comments of


                                            Matteo Merli

Client is highly fork-unsafe: any usage of multiprocessing deadlocks if an at-fork hook triggers garbage collection of a Pulsar client

I believe we would have to inform Asio about the happening fork (https://www.boost.org/doc/libs/1_82_0/doc/html/boost_asio/reference/io_context/notify_fork.html), though I don't see an immediate easy way to hook it up automatically, in the C++ code.

Client is highly fork-unsafe: any usage of multiprocessing deadlocks if an at-fork hook triggers garbage collection of a Pulsar client

I tried to expose the manual hook in C++ client and call it from Python wrapper. So far I was not successful in not making it crash when ### C++...

[pulsar-client-cpp] Excessive locking cause significant performance degradation

> Are you willing accept PR for issue number one or both? @AndrewJD79 Fixes for any of these are warmly welcomed.

[Feature request] Conan recipe to build C++ client and publication to conan-center

@cbornet I don't have any experience with Conan.io but it seems to be a good way to package the client lib. > log4cxx (but it is optional, isn't it ?):...

[Feature request] Conan recipe to build C++ client and publication to conan-center

Oh, I see : https://github.com/apache/pulsar/blob/579e0dc74039c081301662e5fd55209c1a4abe5c/pulsar-client-cpp/lib/checksum/crc32c_sw.cc#L47 It should be easy to just switch to same thing from boost https://www.boost.org/doc/libs/1_68_0/doc/html/thread/synchronization.html#thread.synchronization.once

[Feature request] Conan recipe to build C++ client and publication to conan-center

We actually use that in another place: https://github.com/apache/pulsar/blob/8b2929b3e5403fb44653976dc74be287666f7b96/pulsar-client-cpp/lib/c/c_MessageId.cc#L37

[C++] pulsar-client-cpp stability enhancements and code cleanup

@oversearch All the proposed changes look great!

disableReplication in MessageBuilder does not work in PulsarClient C++ and Java

@tuan6956 It should be `disableReplication(true)` rather than `false`

PIP-144: Making SchemaRegistry implementation configurable

@eolivelli I think we should consider the NAR approach in a separate PR

[pulsar-broker] Broker handle back-pressure with max-pending message across topics to avoid OOM

@rdhabalia there is already a mechanism in place. ptal at #7406