diff options
Diffstat (limited to 'vendor/github.com/jtolds/gls/README.md')
-rw-r--r-- | vendor/github.com/jtolds/gls/README.md | 89 |
1 files changed, 89 insertions, 0 deletions
diff --git a/vendor/github.com/jtolds/gls/README.md b/vendor/github.com/jtolds/gls/README.md new file mode 100644 index 0000000..4ebb692 --- /dev/null +++ b/vendor/github.com/jtolds/gls/README.md @@ -0,0 +1,89 @@ +gls +=== + +Goroutine local storage + +### IMPORTANT NOTE ### + +It is my duty to point you to https://blog.golang.org/context, which is how +Google solves all of the problems you'd perhaps consider using this package +for at scale. + +One downside to Google's approach is that *all* of your functions must have +a new first argument, but after clearing that hurdle everything else is much +better. + +If you aren't interested in this warning, read on. + +### Huhwaht? Why? ### + +Every so often, a thread shows up on the +[golang-nuts](https://groups.google.com/d/forum/golang-nuts) asking for some +form of goroutine-local-storage, or some kind of goroutine id, or some kind of +context. There are a few valid use cases for goroutine-local-storage, one of +the most prominent being log line context. One poster was interested in being +able to log an HTTP request context id in every log line in the same goroutine +as the incoming HTTP request, without having to change every library and +function call he was interested in logging. + +This would be pretty useful. Provided that you could get some kind of +goroutine-local-storage, you could call +[log.SetOutput](http://golang.org/pkg/log/#SetOutput) with your own logging +writer that checks goroutine-local-storage for some context information and +adds that context to your log lines. + +But alas, Andrew Gerrand's typically diplomatic answer to the question of +goroutine-local variables was: + +> We wouldn't even be having this discussion if thread local storage wasn't +> useful. But every feature comes at a cost, and in my opinion the cost of +> threadlocals far outweighs their benefits. They're just not a good fit for +> Go. + +So, yeah, that makes sense. That's a pretty good reason for why the language +won't support a specific and (relatively) unuseful feature that requires some +runtime changes, just for the sake of a little bit of log improvement. + +But does Go require runtime changes? + +### How it works ### + +Go has pretty fantastic introspective and reflective features, but one thing Go +doesn't give you is any kind of access to the stack pointer, or frame pointer, +or goroutine id, or anything contextual about your current stack. It gives you +access to your list of callers, but only along with program counters, which are +fixed at compile time. + +But it does give you the stack. + +So, we define 16 special functions and embed base-16 tags into the stack using +the call order of those 16 functions. Then, we can read our tags back out of +the stack looking at the callers list. + +We then use these tags as an index into a traditional map for implementing +this library. + +### What are people saying? ### + +"Wow, that's horrifying." + +"This is the most terrible thing I have seen in a very long time." + +"Where is it getting a context from? Is this serializing all the requests? +What the heck is the client being bound to? What are these tags? Why does he +need callers? Oh god no. No no no." + +### Docs ### + +Please see the docs at http://godoc.org/github.com/jtolds/gls + +### Related ### + +If you're okay relying on the string format of the current runtime stacktrace +including a unique goroutine id (not guaranteed by the spec or anything, but +very unlikely to change within a Go release), you might be able to squeeze +out a bit more performance by using this similar library, inspired by some +code Brad Fitzpatrick wrote for debugging his HTTP/2 library: +https://github.com/tylerb/gls (in contrast, jtolds/gls doesn't require +any knowledge of the string format of the runtime stacktrace, which +probably adds unnecessary overhead). |