Prefilter graphs in the uploader #3497
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
caisq
reviewed
Apr 9, 2020
|
|
||
|
|
||
| def migrate_event(event): | ||
| def migrate_event(event, experimental_filter_graph=False): |
Contributor
There was a problem hiding this comment.
Please add doc string for the new kwarg.
| ValueError: If `large_attrs_key is None` while `limit_attr_size != None`. | ||
| ValueError: If `limit_attr_size` is defined, but <= 0. | ||
| """ | ||
| # TODO(@davidsoergel): detect whether a graph has been filtered already |
Contributor
There was a problem hiding this comment.
To make this TODO item clearer, I believe you can say in addition: "if it is already filtered, return immediately".
tensorboard/dataclass_compat_test.py
Outdated
|
|
||
| def test_graph_def_experimental_filter_graph(self): | ||
| # Create a `GraphDef` and write it to disk as an event. | ||
| logdir = self.get_temp_dir() |
Contributor
There was a problem hiding this comment.
Question about testing strategy: is it really necessary to create a logdir and a writer? If your intention here is to simply test self._migrate_event() (and ultimately dataclass_compat.migrate_event()), you can just construct a Event, pass it to that method and check the result, right?
Member
Author
There was a problem hiding this comment.
Oh good point. I had pasted that from the test above but here the logdir part is superfluous. (There too, maybe, but that's out of scope). Thanks!
caisq
approved these changes
Apr 9, 2020
Contributor
caisq
left a comment
caisq
left a comment
There was a problem hiding this comment.
LGTM. Thanks for reducing the size of the uploaded graphs.
davidsoergel
changed the title
davidsoergel
added a commit
that referenced
this pull request
Apr 10, 2020
Since #3497, we parse GraphDefs in dataclass_compat.py during upload. If a graph is corrupt, that parsing fails. Here we catch the resulting exception, issue a warning, and continue (omitting the graph). This also updates tests to use valid GraphDefs where appropriate, as opposed to bytes(1024), which apparently produces inconsistent results with different proto parsers (e.g., OSS vs. Google internal).
bileschi
pushed a commit
to bileschi/tensorboard
that referenced
this pull request
Apr 15, 2020
There is no point in uploading GraphDef data that won't be displayed anyway. In particular, the TensorBoard frontend filters out node attributes larger that 1024 bytes, since it has no good way to present those. So we may as well filter those out before upload to TensorBoard.dev, so as not to waste bandwidth, storage, and read-time processing.
bileschi
pushed a commit
to bileschi/tensorboard
that referenced
this pull request
Apr 15, 2020
Since tensorflow#3497, we parse GraphDefs in dataclass_compat.py during upload. If a graph is corrupt, that parsing fails. Here we catch the resulting exception, issue a warning, and continue (omitting the graph). This also updates tests to use valid GraphDefs where appropriate, as opposed to bytes(1024), which apparently produces inconsistent results with different proto parsers (e.g., OSS vs. Google internal).
bileschi
pushed a commit
that referenced
this pull request
Apr 15, 2020
There is no point in uploading GraphDef data that won't be displayed anyway. In particular, the TensorBoard frontend filters out node attributes larger that 1024 bytes, since it has no good way to present those. So we may as well filter those out before upload to TensorBoard.dev, so as not to waste bandwidth, storage, and read-time processing.
bileschi
pushed a commit
that referenced
this pull request
Apr 15, 2020
Since #3497, we parse GraphDefs in dataclass_compat.py during upload. If a graph is corrupt, that parsing fails. Here we catch the resulting exception, issue a warning, and continue (omitting the graph). This also updates tests to use valid GraphDefs where appropriate, as opposed to bytes(1024), which apparently produces inconsistent results with different proto parsers (e.g., OSS vs. Google internal).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There is no point in uploading GraphDef data that won't be displayed anyway. In particular, the TensorBoard frontend filters out node attributes larger that 1024 bytes, since it has no good way to present those. So we may as well filter those out before upload to TensorBoard.dev, so as not to waste bandwidth, storage, and read-time processing.