ghophp / buildbot-dashboard Goto Github PK

the fetchers stop after a while, this is the log we got after hours running:

2015-12-02_15:59:08.10428 [buildbot-dashboard] fetching builders at 25 goroutines
2015-12-02_16:00:08.10636 [sockets] [Error] [10.100.107.243:58345] Error writing to socket: write tcp 10.0.0.5:8080->10.100.107.243:58345: i/o timeout
2015-12-02_16:00:08.10641 [sockets] [Error] [10.100.107.243:58345] Error pinging socket: write tcp 10.0.0.5:8080->10.100.107.243:58345: use of closed network connection
2015-12-02_16:00:08.10647 [sockets] [Error] [10.100.107.243:58345] Connection could not be closed: close tcp 10.0.0.5:8080->10.100.107.243:58345: use of closed network connection
2015-12-02_16:00:08.10652 [sockets] [Error] [10.100.107.243:58344] Error reading from socket: read tcp 10.0.0.5:8080->10.100.107.243:58344: i/o timeout
2015-12-02_16:00:08.10659 [sockets] [Error] [10.100.107.243:58344] Error writing to socket: write tcp 10.0.0.5:8080->10.100.107.243:58344: i/o timeout
2015-12-02_16:00:08.10662 [sockets] [Error] [10.100.107.243:58344] Error pinging socket: write tcp 10.0.0.5:8080->10.100.107.243:58344: use of closed network connection
2015-12-02_16:00:08.10666 [sockets] [Error] [10.100.107.243:58344] Connection could not be closed: close tcp 10.0.0.5:8080->10.100.107.243:58344: use of closed network connection
2015-12-02_16:00:08.10670 [sockets] [Error] [10.100.107.243:58345] Error reading from socket: read tcp 10.0.0.5:8080->10.100.107.243:58345: use of closed network connection

this means that the problem is not from goroutines number, but something related to ws.. we maybe must extend the logs for this sockets, and check why is it hanging forever in there..

if there are multiple HTTP requests to fetch the builder there is a possible overhead on the server, we can do a throughput style with cache, just use the same cache structure that is on the fetchBuilders method.

travis offer multiple envs, maybe we can use that to create the precompiled binaries on the CI and then submit to the s3 and link with a release..

there is much more tests that can be written to the backend and the frontend.. the backend is priority and the frontend can be done together with the issue #4

As you can see below the dashboard dont work really well in a 21:9 monitor
http://imgur.com/gallery/no3t9ib

buildbot start the builds from #0, we need to differentiate from no builds..

the project works fine with zoom.. maybe we should consider to remove the size flag.. it is not necessary.. this should be done maybe with the firefox and safari bug load, and we need to check if all the browsers (important ones) are fine with the zoom..

need to implement a build process on js to output a single file there.. maybe use this jsx also to write better react code..

I'm using buildbot:latest from docker and i'm unable to make it work, it looks that the endpoints aren't working and as described in the buildbot page:

The API described here is version 2. The ad-hoc API from Buildbot-0.8.x, version 1, is no longer supported.

Is that the case?

create a table as the go lang download page, that adds the compressed binaries and the checksums.. this gives more reliability to the project..

https://github.com/ghophp/buildbot-dashboard/blob/master/handler/builders.go#L136

golang allow direct removal..

https://github.com/ghophp/buildbot-dashboard/blob/master/handler/builders.go#L56

remove the timeout that is not being used

after a while the ws just die, we need to implement a log library in order to track this problem down..

when we remove a builder from the server, we need to remove it from the cache and the arrangement..

the cachedBuilds could be empty, but the project has builds, instead of use that, it is better to look for the last number in the last build..

the websocket could lost connection for any reason, so it would be good to have a reconnect process on the javascript side..

maybe submit the project to the ports of freebsd do allow easy install.. the same for brew and apt-get.. if possible..

on ff and safari if your buildbot has a lot of builders (eg. apache 270~), the browser freezes until load.. need to investigate the cause of this problem..

when a new builder is added, we should add him on-the-fly when the socket send a new id.. we can wait the cache to refresh, no problem.. but when the key is sent over the websocket, we must add it right the way:

https://github.com/ghophp/buildbot-dashboard/blob/master/static/assets/js/app.js#L66

the states need a better recognition process.. we can iterate over the texts and see if we find any of the key values... that is most appropriate way to do it..

https://github.com/ghophp/buildbot-dashboard/blob/master/handler/builders.go#L77

we must include a button or something in the interface to allow a request on /builders that send a parameters that force the cache reload.. this will avoid cache issues.. and maybe can also solve the remove problem..

create a build process for the CI that will bind the zip to the final binary.. also need to change the code to consume the assets in this way

there is a race condition on the process of fetching and sending as broadcast since the last commit to improve the performance.. I will have to work to improve this.. and present a better socket structure..

after a long period, the system just stop to send messages.. it still respond to the /builders and the ws connections, so we need to assume that there is something to do with the routines to fetch the builders, we need to investiga this with the #23 feature..