tokopedia / logging

common Go logging

logging

common logging/tracing package for go servers

API documentation: https://godoc.org/gopkg.in/tokopedia/logging.v1

logging's Issues

Could not found API documentation

I tried to visit the docs : https://godoc.org/gopkg.in/tokopedia.com/logging.v1 but not found.

I was kinda curious what's the different between package log in go and this one?

"-version" flag (and it's handler) won't work

To start from my understanding, it seems that tokopedia/logging tries to define its own buildhash build time info to get build information

logging/init.go

Line 36 in 41f21cd

flag.BoolVar(&versionFlag, "version", false, "binary version")

logging/init.go

Lines 60 to 63 in 41f21cd

	if versionFlag {
	fmt.Println(appVersion())
	os.Exit(0)
	}

logging/version.go

Lines 1 to 7 in 41f21cd

	package logging
	var buildhash string
	func appVersion() string {
	return buildhash
	}

This won't work, since buildhash is injected during build time, so it could only be used on package main

as per build_debian_package roles
https://github.com/tokopedia/ansible-v2/blob/26ac51e1695816ad771b9e75c945cdda76e5313a/roles/build_debian_package/templates/debian-rules-file.j2#L77

doesn't play well with socketmaster

When we use socketmaster- the SIGHUP is sent to the socketmaster binary, which then forks a new child process. It then sends a SIGTERM to the old process. So the rotation code is never executed, because SIGHUP is never sent.

Thats not it - for some reason, the newly started binary actually has the wrong file opened as stderr, so the error logs keep going to the old file.

The solution could be to modify socketmaster, to use -USR1 signal for restarting the process, and making the -HUP passthrough, but there is still a problem.How do we indicate to upstart when we want an app reload and when we intend only a logrotation.

[IDEA] Closing stderr and stdout in sigterm

We couldn't use socketmaster reload for logrotate, this is caused by 1 and 2 fd is not closed since the fact that the main process (which is socketmaster) doesn't get terminated.
Socketmaster will send SIGTERM to the app on reload (SIGHUP to socketmaster), then we can add handler to close the 1 and 2 fd at that time.
So we don't have to restart the app every log rotate.

Do you think it's possible?

(still in the long week end vacation, will take a look by myself at monday)