[Maestro Sink] Double init

[chhaine]

What happened?

We noticed in some experiment the Sink constructor is called twice. I do not know if this is intended behaviour, looking at other sinks it seems reasonably harmless, however calling mstro_init (call located in the MaestroSink constructor) twice is a bug.

What are the steps to reproduce the bug?

We do not know how or why the Sink constructor is called twice, and rely on your expertise.

Perhaps @AOA-Mohammed can give a bit more info?

Version

v2.2.2

Platform (OS and architecture)

(Marenostrum5 GPP compute nodes) Linux gs22r1b21 5.14.0-284.30.1.el9_2.x86_64 #1 SMP PREEMPT_DYNAMIC Fri Aug 25 09:13:12 EDT 2023 x86_64 x86_64 x86_64 GNU/Linux

Relevant log output

I don't think we can see any useful log on the MultIO side, the Maestro side shows the following error message:


[E:core] SIM_1 -- 1748:0 3 t22725235404992 (gs29r3b18 3747855) 3043131169558870: mstro_core_init(core.c:546) Init called twice without finalization in-between.


@AOA-Mohammed

Accompanying data

No response

Organisation

Hewlett Packard Enterprise

[AOA-Mohammed]

Thanks @chhaine for reporting this.
At a high level, I see the issue is that MultIO uses a factory pattern to build sinks, i.e., DataSinkBuilder.
While Maestro supports the use of multiple applications/user threads, it only allows them to be initialized once from a single process. That is, a single thread can initialize Maestro, then multiple threads within the same process can make Maestro calls.

I am not exactly sure what configuration led to having MultIO building multiple MaestroSinks, but I think the MultIO code around building data sinks may need to handle MaestroSink type a bit differently.
Maybe the simplest fix would be to implement a guard in the constructor of the MaestroSink object to avoid initializing Maestro multiple times within a process.

[AOA-Mohammed]

I have created #227 as a proposal to fix this issue, basically allowing multio to handle maestro "Not twice" status as not error.