-
Notifications
You must be signed in to change notification settings - Fork 9
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Looks entirely plausible to me from an ops perspective; adding Kelly to check whether these Dockerfile changes have broken anything on the dev side.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
@@ -46,7 +33,7 @@ RUN mkdir -p ${HOME}/lib/ ${HOME}/bin/ ${HOME}/data/models/ ${HOME}/src/ds-srv/ | |||
|
|||
RUN curl https://sh.rustup.rs -sSf | sh -s -- -y --default-toolchain stable | |||
|
|||
RUN curl https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.v${DS_VER}.gpu/artifacts/public/native_client.tar.xz -sSL | pixz -d | tar -C ${HOME}/lib/ -xf - | |||
RUN curl https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.v${DS_VER}.gpu/artifacts/public/native_client.tar.xz -sSL | xz -d | tar -C ${HOME}/lib/ -xf - | |||
|
|||
RUN curl https://github.com/mozilla/DeepSpeech/releases/download/v${DS_VER}/deepspeech-${DS_VER}-models.tar.gz -sSL | gunzip | tar -C ${HOME}/data/models/ --strip-components 1 -xvf - |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
For now this is OK, but eventually we will want to have the ability to have custom models, one for Firefox Reality, one for Firefox Voice....
@@ -46,7 +33,7 @@ RUN mkdir -p ${HOME}/lib/ ${HOME}/bin/ ${HOME}/data/models/ ${HOME}/src/ds-srv/ | |||
|
|||
RUN curl https://sh.rustup.rs -sSf | sh -s -- -y --default-toolchain stable | |||
|
|||
RUN curl https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.v${DS_VER}.gpu/artifacts/public/native_client.tar.xz -sSL | pixz -d | tar -C ${HOME}/lib/ -xf - | |||
RUN curl https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.v${DS_VER}.gpu/artifacts/public/native_client.tar.xz -sSL | xz -d | tar -C ${HOME}/lib/ -xf - |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Not sure how much this is a problem ? But it looks good
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
pixz doesn't support decompression to stdout, so this docker step wasn't putting these files in the docker image
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That would be unfortunate, since I'm using pixz -d
to stdout in a lot of places on TaskCluster for DeepSpeech CI.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
$ ll lib/
total 0
drwxr-xr-x 1 alex alex 0 mars 4 18:06 .
drwxr-xr-x 1 alex alex 432 mars 4 18:06 ..
$ pixz --version
pixz: invalid option -- '-'
pixz: Parallel Indexing XZ compression, fully compatible with XZ
Basic usage:
pixz input output.pxz # Compress a file in parallel
pixz -d input.pxz output # Decompress
Tarballs:
pixz input.tar output.tpxz # Compress and index a tarball
pixz -d input.tpxz output.tar # Decompress
pixz -l input.tpxz # List tarball contents very fast
pixz -x path/to/file < input.tpxz | tar x # Extract one file very fast
tar -Ipixz -cf output.tpxz dir # Make tar use pixz automatically
Input and output:
pixz < input > output.pxz # Same as `pixz input output.pxz`
pixz -i input -o output.pxz # Ditto
pixz [-d] input # Automatically choose output filename
Other flags:
-0, -1 ... -9 Set compression level, from fastest to strongest
-p NUM Use a maximum of NUM CPU-intensive threads
-t Don't assume input is in tar format
-k Keep original input (do not remove it)
-c ignored -h Print this help
pixz 1.0.6
(C) 2009-2012 Dave Vasilevsky <[email protected]>
https://github.com/vasi/pixz
You may use this software under the FreeBSD License
$ curl https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.v0.6.1.gpu/artifacts/public/native_client.tar.xz -sSL | pixz -d | tar -C lib/ -xf -
can not seek in input: Illegal seek
alex@portable-alex:~/tmp/deepspeech/models$ ll lib/
total 72M
drwxr-xr-x 1 alex alex 144 mars 4 18:08 .
drwxr-xr-x 1 alex alex 432 mars 4 18:06 ..
-rwxr-xr-x 1 alex alex 892K janv. 10 18:22 deepspeech
-rw-r--r-- 1 alex alex 8,4K janv. 10 18:17 deepspeech.h
-r-xr-xr-x 1 alex alex 2,1M janv. 10 18:19 generate_trie
-r-xr-xr-x 1 alex alex 69M janv. 10 18:22 libdeepspeech.so
-rw-r--r-- 1 alex alex 17K janv. 10 18:17 LICENSE
-rw-r--r-- 1 alex alex 1,2K janv. 10 18:17 README.mozilla
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@jbuck So, indeed, it does support it. I'm worried that you see that kind of divergence.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Ah, I was assuming it was a fatal error when I got this error while running the docker build:
Step 6/7 : RUN curl https://community-tc.services.mozilla.com/api/index/v1/task/project.deepspeech.deepspeech.native_client.v${DS_VER}.gpu/artifacts/public/native_client.tar.xz -sSL | pixz -d | tar -C ${HOME}/lib/ -xf -
---> Running in 4e8e178a20ad
can not seek in input: Illegal seek
Removing intermediate container 4e8e178a20ad
---> 1e097ede4bd9
But it did decompress the files successfully inside the docker image
It looks like this is filed as a bug on the pixz repo: vasi/pixz#62
No description provided.