Webgl racecondition fix #440
Conversation
|
Looks reasonable, although I have not been able to replicate the issue the sleeps are fixing. Looks like all these lines could be removed, and other sleeps in |
|
I tested this change on my aging macbook air, and it works fine. Re the sleeps: I was hoping they would be helpful in solving the (random) webgl crashes I would get from time to time (see the trace below—I'm not 100% sure, but it seems like webgl is taking too long to set the new attributes?). $ python multi_panels_plots.py
Started server on port 3266
{'camera.azimuth': 0, 'camera.altitude': 90, 'camera.target': [0, 0, 0], 'surface.{subject}.unfold': 0.5, 'surface.{subject}.pivot': 180, 'surface.{subject}.shift': 10, 'surface.{subject}.specularity': 0, 'surface.{subject}.sampler': 'nearest', 'surface.{subject}.layers': 1}
{'camera.azimuth': 180, 'camera.altitude': 90, 'camera.target': [0, 0, 0], 'surface.{subject}.unfold': 0.5, 'surface.{subject}.pivot': 180, 'surface.{subject}.shift': 10, 'surface.{subject}.specularity': 0, 'surface.{subject}.sampler': 'nearest', 'surface.{subject}.layers': 1}
Traceback (most recent call last):
File "multi_panels_plots.py", line 42, in <module>
cortex.export.plot_panels(volume, **params)
File "/Users/mvdoc/Documents/repos/pycortex/cortex/export/panels.py", line 108, in plot_panels
layers=layers,
File "/Users/mvdoc/Documents/repos/pycortex/cortex/export/save_views.py", line 114, in save_3d_views
handle._set_view(**this_view_params)
File "/Users/mvdoc/Documents/repos/pycortex/cortex/webgl/view.py", line 488, in _set_view
subject_list = self.ui.surface._folders.attrs.keys()
File "/Users/mvdoc/Documents/repos/pycortex/cortex/webgl/serve.py", line 328, in __getattr__
assert attr in self.attrs
AssertionError |
|
@mvdoc that's very closely related to the problem this branch is fixing, which was also intermittent and related to those javascript-derived variables being filled with values via the websocket. Have you tried replicating that with |
|
Yes, I still hit the same issue (randomly) with the |
mvdoc
left a comment
There was a problem hiding this comment.
I think this PR can be merged. It didn't solve the issue I often hit, but the change in the code seems reasonable (writing with async). I would maybe keep the sleeps in there just as a safety measure.
In somewhat related things, there's an annoying Heisen-bug in how the tornado websocket communication works. If it tries to send too many queries too quickly, tornado just barfs. This fix is based on: tornadoweb/tornado#2871
I believe this fix works well — I haven't been able to replicate the problem since applying this fix — but the problem was random before, so I'm not 100% certain.
I also believe that this fix should make a lot of
time.sleepcalls unnecessary (removing them seemed to have no effect), but, again, the effect of those pauses is inconsistent.