Serving: Error at model server startup if logging_config is set

[giavjeko]

Model server config (text formatted protobuf):

model_config_list {
  config {
    name: "default"
    base_path: "/data/model"
    model_platform: "tensorflow"
    version_policy: ALL_VERSIONS
    logging_config {
      log_collector_config {
        type: "",
        filename_prefix: "/log/model-server"
      }
      sampling_config {
        sampling_rate: 1
      }
    }
  }
}

Startup fails with the following error:

F tensorflow_serving/model_servers/main.cc:445] Check failed: ::tensorflow::Status::OK() == (ServerCore::Create(std::move(options), &core)) (OK vs. Invalid argument: No request-logger-creator provided.)

[sukritiramesh]

From a quick check, it looks like it attempts to perform an update without access to a request logger creator when the logging config map is not empty.

@vinuraja, is this expected?

[giavjeko]

This still happens in r0.6 branch. Any info?

[tvkamara]

I also stumbled across this error. I'm just trying to get some logging out of the sample model server. Specifically to log the requests and responses.

[giavjeko]

This still happens in r1.5 branch. Any info?

[vinuraja]

Thanks for reporting this. Sorry, this is expected at this point in time. We haven't yet added full support for logging requests to the model-server. We are having design discussions on this within the team, and will get back to you.

[Harshini-Gadige]

@vinuraja Any update on this ?

[mraduldubey]

Please provide an update on this. Thanks.

[mcf330]

Is there any workaround to log the requests rather than enable the TF_CPP_MIN_VLPG_LEVEL flag? thanks.

[mikezhang95]

any updates on this?