DTrace Hangout 19 January 2017
Domagoj Stolfa, James McIlree, George Neville-Neil
- gnn has added an rfd repo to the OpenDTrace organization on github, and proposes that we use the RFD process, started by Joyent to discuss changes to DTrace. RFD 0 is "how to write and RFD" and it is likely that RFD 1 will be the description of copyoutmbuf and copyoutmbufat.
- gnn has written two subroutines to handle copyoutmbuf and copyoutmbufat which will take all the packet data from an mbuf chain and return it to a caller in D. That code is currently in a private repo but will be shared shortly.
- gnn has written a first draft of a DTrace specification which be shared at a later point
- gnn is proposing a "true" if statement replace the syntactic sugar version that we currently have.
- Discussion of loops. James mentions that loops are one of the places that the original design tensions come into play. The first version of loops will be bounded, unless in destructive.
Domagoj wants to add a fifth tuple entry which is an instance and resides at the front of the probe. This will become a new RFD as well.
- There is an issue with any type of distributed tracing, including with a bunch VMs, because where we maintain the mappings is important. Right now they're in the kernel, but that's not scalable or right for a distributed case.
Arun mentions that BSDCan is coming up in June (https://www.bsdcan.org/2017/) and that it would be good for us to meet there.