optimize Talos core components memory usage

[smira]

From a worker node:

apid is 80MiB, dashboard is 94MiB and init is 128 MiB

COSI Improvements

Preview Give feedback

1. rework the controller dependency database cosi-project/runtime#305
   +0
   
2. optimize metadata.Equal cosi-project/runtime#306
   +0
   
3. run pluralize not in the run time cosi-project/runtime#307
   +0

3rd party libraries

Preview Give feedback

1. refactor: optimize port map to switch statement gopacket/gopacket#24
   +0
   
2. refactor: don't fill empty metadata slots gopacket/gopacket#32
   +0
   

Talos Improvements

Preview Give feedback

1. use a shared gRPC buffer pool on the receive path #7576
   +0
2. optimize the manifest conversion from bytes to unstructured #7577
   +0
3. optimize init memory allocations #7578
   +0
4. reimplement packet capture in machined #7880
   +0
   

[smira]

There was some work done recently to optimize memory usage, but there's still some room for improvement here

[smira]

Things got better with merged in fixes, but still a lot of room for improvement:

[smira]

• big text section size (code, read-only, shared between Talos processes)
• big gopclntab section size (used for stacktraces, read-only, shared between Talos processes)
• RSS in dashboard doesn't show the real size of the process, as it doesn't account for text/gopclntab shared across processes
• readelf -a -W _out/machined provides an overview of section sizes and mapping into memory
• gsv wraps bloaty to show size per module/package (code & data)
• Go heap/stack size

[smira]

dropping gopacket library from machined reduces observed RSS down ~20 MiB