Are there any simple examples/documentation on multi-machine tracing?

[cphlipot1]

Is there any documentation or simple examples of how to use traced_relay to collect multi machine traces?

Likewise, is there any documentation or examples of valid traces utilizing multiple machine ids for those wishing to synthetically create similar traces in situations where traced_relay can't be used directly?

[LalitMaganti]

Is there any documentation or simple examples of how to use traced_relay to collect multi machine traces?

This is not unfortunately. It's very optimised for ChromeOS and VM tracing atm, I'm not sure it's in a state anyone else can use.

Likewise, is there any documentation or examples of valid traces utilizing multiple machine ids for those wishing to synthetically create similar traces in situations where traced_relay can't be used directly?

The correct answer for this is to wait for #1018. Writing multi-machine traces inside a single file is very complicated and there are subtle dependencies baked in which I would not recommend anyone to do. The much better option is to write each trace separately and "merge" the traces together in the UI/trace_processor.

In the coming month, I will implement support for merging multiple traces for the same device. Once we do that, it is a pretty small jump to implement multiple traces from multiple devices.

[cphlipot1]

Thanks for the context. I will follow along in #1018.

Can you clarify if the intent is to build a single trace containing data from multiple machines, or if it is more about allowing trace processor to support multiple trace files simultaneously in a single session?

[LalitMaganti]

More the latter. Though that will still be a "trace" in the eyes of trace processor, just containing multiple independent streams of data.

If you know video formats, think of it like MKV: it's a container format that can contain multiple independent streams. All the MKV is, is some metadata which puts that together

I'm looking to do something similar except with zip files + a JSON manifest.

[cphlipot1]

If you know video formats, think of it like MKV: it's a container format that can contain multiple independent streams. All the MKV is, is some metadata which puts that together

thanks that makes sense.

I'm looking to do something similar except with zip files + a JSON manifest.

i had a similar idea of ZIP + JSON metadata, for how we might load traces from multiple machines even if the trace was a non-protobuf format trace and/or that did not have sufficient data to convey machine ids or clock offsets. (eg. chrome-json traces, perf.data, dumpstate.txt, etc.)

Is that something you think your ZIP + JSON manifest solution you had in mind would be flexible enough to handle?

[LalitMaganti]

Yeah absolutely. All the information about clock syncing etc should live in the JSON manifest

To be clear, my first priority is getting it working on traces collected on the same machine. Only later will I deal with multiple machines.

[cphlipot1]

Makes sense. Thank you for sharing the info. That seems like a sensible approach. I'll keep eye out for when this is ready to use.