Andy Brown

Compression on the relay isn't the problem. The java client is compressing a metrics batch. The relay decompresses and forwards to carbon cache. The problem is that the whole compressed...

Looking briefly at the source code for both the relay and the lz4 library it seems the problem is the use of the block streaming API by the relay. `LZ4_decompress_safe_continue()`...

> I just tested, using two relays, it works as expected. The problem appears to be that nothing seems to be able to just generate block streams, except the c-library...

Sorry for the late reply, I must have missed the notification for this. zstd looks almost too good to be true. We'd certainly switch over to that if it were...

I'll have a look at the compression performance of Snappy against our payloads. I have a feeling it will be good because our payloads contain a lot of repeating text...

I tested some of our payloads and found that with zlib set to level 5 compression we got 95.5% and with snappy it's about 90% using the 'snzip' command line...

Would you be interested in a pull request that implements the LZ4 framed format? I've got it working here and have tested communication between java -> relay and relay ->...

Cool, I'll leave it running for a bit longer capturing metrics and forwarding between relays to make sure it's all good then work on the PR. I'll branch off your...

Thank you very much for contributing. I tried to build your PR from the command line and got lots of errors. This was the first: ``` $ scons mode=debug mcu=f401...

> > > I have the examples built correctly, but haven't pushed my changes yet, because I'd like to ask you Andy for your preference. > The system files for...