tensorboard
tensorboard copied to clipboard
tensorflow
Pinned graphs of deleted data can't unpin
Pinned graphs are not cleaned up when working on new log directory or after restarting server.
Screenshot shows the issue. I get the error about hitting max limit of pins, however there are no graphs in pinned section, so i cannot unpin them. This happened because I pinned some graphs, then deleted the log directory, re-started my training process but with newly named graphs. This means the memory of pinned graphs is still in the server somewhere, but i cannot unpin them now. I've tried restarting the tensorboard server. I guess i need to delete some state files but i don't know where to find them.
Hi @mattthelee,
Can you please try killing and restating the current TensorBoard process (as described in https://stackoverflow.com/questions/36896164/tensorflow-how-to-close-tensorboard-server)?
I killed it by doing ctrl + c which appears to have killed the server. Are you saying there is a background process running as well?
Hi @mattthelee,
Yes, occasionally CTRL + C doesn't suffice. Can you please look up the TensorBoard process and make sure it's killed before restarting another instance? Otherwise it will attempt to reuse the existing process.
Sorry for the delayed response: As far as I know, the only place where the list of pinned graphs is stored is in the app's memory and the URL. This does not get stored on the server. Loading the page without any query parameters should fix the problem.
Having said that, it would be nice if we handled this a bit more gracefully. Perhaps removing the pinned graphs from the URL when we can conclude that the corresponding tags or runs no longer exist.
