Mkfifo and dx cat
Last updated
Was this helpful?
Last updated
Was this helpful?
This applet performs a SAMtools count on an input file while minimizing disk usage. For additional details on using FIFO (named pipes) special files, run the command man fifo in your shell.Warning: Named pipes require BOTH a stdin and stdout or they will block a process. In these examples, we place incomplete named pipes in background processes so the foreground script process does not block.
To approach this use case, let’s focus on what we want our applet to do:
Stream the BAM file from the platform to a worker.
As the BAM is streamed, count the number of reads present.
Output the result into a file.
Stream the result file to the platform.
First, we establish a named pipe on the worker. Then, we stream to stdin of the named pipe and download the file as a stream from the platform using .
First, we establish a named pipe on the worker. Then, we stream to stdin of the named pipe and download the file as a stream from the platform using .
FIFO
stdin
stdout
BAM file
YES
NO
Now that we have created our FIFO special file representing the streamed BAM, we can just call the samtools command as we normally would. The samtools command reading the BAM would provide our BAM FIFO file with a stdout. However, keep in mind that we want to stream the output back to the platform. We must create a named pipe representing our output file too.
FIFO
stdin
stdout
BAM file
YES
YES
output file
YES
NO
The directory structure created here (~/out/counts_txt) is required to use the dx-upload-all-outputs command in the next step. All files found in the path ~/out/<output name> will be uploaded to the corresponding <output name> specified in the dxapp.json.
Currently, we’ve established a stream from the platform, piped the stream into a samtools command, and finally outputting the results to another named pipe. However, our background process is still blocked since we lack a stdout for our output file. Luckily, creating an upload stream to the platform will resolve this.
We can upload as a stream to the platform using the commands dx-upload-all-outputs or dx upload -. Make sure to specify --buffer-size if needed.
FIFO
stdin
stdout
BAM file
YES
YES
output file
YES
YES
Note: Alternatively, dx upload - can upload directly from stdin. In this example, we would no longer need to have the directory structure required for dx-upload-all-outputs.Warning: When uploading a file that exists on disk, dx upload is aware of the file size and automatically handles any cloud service provider upload chunk requirements. When uploading as a stream, the file size is not automatically known and dx upload uses default parameters. While these parameters are fine for most use cases, you may need to specify upload part size with the --buffer-size option.
Now that our background processes are no longer blocking the rest of the applet’s execution, we simply wait in the foreground for those processes to finish.
Note: If we didn’t wait the app script would running in the foreground would finish and terminate the job! We wouldn’t want that.
The SAMtools compiled binary is placed directly in the <applet dir>/resources directory. Any files found in the resources/ directory will be uploaded so that they will be present in the worker’s root directory. In our case:
When this applet is run on a worker, the resources/ folder will be placed in the worker’s root directory /:
/usr/bin is part of the $PATH variable, so we can reference the samtools command directly in our script as samtools view -c ...
The directory structure created here (~/out/counts_txt) is required to use the command in the next step. All files found in the path ~/out/<output name> will be uploaded to the corresponding <output name> specified in the dxapp.json.
We can upload as a stream to the platform using the commands or . Make sure to specify --buffer-size if needed.