CODEWITHSUNDEEP
tensorflowtensorboard
MrCartoonology
MrCartoonology

Reputation: 2067

tensorboard not displaying anything

The nodes that I have to work on don't have many outside internet connections. When on one of these nodes, after running tensorboard and visiting http://localhost:6006/, I don't see anything - just a blank page - for a long time, finally I get something.

On a node with internet access, things came up faster. Is tensorboard going to outside URL's for styling and fonts and things like that? On the internal node, I see all these messages that have /external in them, like

27.0.0.1 - - [21/Jan/2016 13:06:05] "GET /external/iron-meta/iron-meta.html HTTP/1.1" 200 -
127.0.0.1 - - [21/Jan/2016 13:06:05] "GET /external/iron-iconset-svg/iron-iconset-svg.html HTTP/1.1" 200 -
127.0.0.1 - - [21/Jan/2016 13:06:05] "GET /external/iron-selector/iron-selection.html HTTP/1.1" 200 -
1

There are also warnings like

WARNING:root:IOError [Errno 2] No such file or directory: '/home/user/anaconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/tensorboard/lib/svg/summary-icon.svg' on path /home/user/anaconda2/envs/tensorflow/lib/python2.7/site-packages/tensorflow/tensorboard/lib/svg/summary-icon.svg

I don't see any simple way to force tensorboard to stay local - or do you think I am barking up the wrong tree here? Maybe there is some other problem, but I didn't get those messages on the internet connected node.

Upvotes: 4

Views: 10768

Answers (2)

pancho33
pancho33

Reputation: 78

Be sure that you include the full path when starting TensorBoard

 tensorboard --logdir='./somedir/logdir'

I´m using safari, crome and firefox without problems to connect to localhost:6006

Upvotes: 1

dandelion
dandelion

Reputation: 1782

TensorBoard does try to load font data externally (it makes a GET request here: https://fonts.googleapis.com/css?family=Roboto:400,300,300italic,400italic,500,500italic,700,700italic ). However, if the font request fails, it shouldn't block the rest of the page from loading. The /external/ refers to external to the TensorFlow repository at build time, but it is all cloned locally.

The summary svg issue suggests to me that you may be using an old version of TensorBoard, although that shouldn't matter for this issue.

I guess that there is some other issue on your non-connected nodes that is causing the initial TensorBoard asset load (which loads a large html file, and a lot of small open-source assets) to be slow. I recommend trying using Google Chrome if you are not already, I have seen issues with Firefox being slow because it loads all the assets in sequence rather than in parallel. If you are using Chrome, try investigating the "network" pane in the Chrome Debugger and reply back with what you find.

Upvotes: 0

Related Questions