Jobs - cortex

Get the Batch API's endpoint
1
cortex get <batch_api_name>
Copied!
Submit a Job
There are three options for providing the dataset for your job:
1.​Data in the request​
2.​List S3 file paths​
3.​Newline delimited JSON file(s) in S3​
Data in the request
The input data for your job can be included directly in your job submission request by specifying an item_list in your json request payload. Each item can be any type (object, list, string, etc.) and is treated as a single sample. item_list.batch_size specifies how many items to include in a single batch.
Each batch must be smaller than 256 KiB, and the total request size must be less than 10 MiB. If you want to submit more data, explore the other job submission methods.
Submitting data in the request can be useful in the following scenarios:
the request only has a few items
each item in the request is small (e.g. urls to images/videos)
you want to avoid using S3 as an intermediate storage layer
1
POST <batch_api_endpoint>:
2
{
3
    "workers": <int>,         # the number of workers to allocate for this job (required)
4
    "timeout": <int>,         # duration in seconds since the submission of a job before it is terminated (optional)
5
    "sqs_dead_letter_queue": {      # specify a queue to redirect failed batches (optional)
6
        "arn": <string>,            # arn of dead letter queue e.g. arn:aws:sqs:us-west-2:123456789:failed.fifo
7
        "max_receive_count": <int>  # number of a times a batch is allowed to be handled by a worker before it is considered to be failed and transferred to the dead letter queue (must be >= 1)
8
    },
9
    "item_list": {
10
        "items": [            # a list items that can be of any type (required)
11
            <any>,
12
            <any>
13
        ],
14
        "batch_size": <int>,  # the number of items per batch (the handle_batch() function is called once per batch) (required)
15
    }
16
    "config": {               # arbitrary input for this specific job (optional)
17
        "string": <any>
18
    }
19
}
20
​
21
RESPONSE:
22
{
23
    "job_id": <string>,
24
    "api_name": <string>,
25
    "kind": "BatchAPI",
26
    "workers": <int>,
27
    "config": {<string>: <any>},
28
    "api_id": <string>,
29
    "sqs_url": <string>,
30
    "timeout": <int>,
31
    "sqs_dead_letter_queue": {
32
        "arn": <string>,
33
        "max_receive_count": <int>
34
    },
35
    "created_time": <string>
36
}
Copied!
The entire job specification is written to /cortex/spec/job.json in the API containers.
S3 file paths
If your input data is a list of files such as images/videos in an S3 directory, you can define file_path_lister in your submission request payload. You can use file_path_lister.s3_paths to specify a list of files or prefixes, and file_path_lister.includes and/or file_path_lister.excludes to remove unwanted files. The S3 file paths will be aggregated into batches of size file_path_lister.batch_size. To learn more about fine-grained S3 file filtering see filtering files.
The total size of a batch must be less than 256 KiB.
This submission pattern can be useful in the following scenarios:
you have a list of images/videos in an S3 directory
each S3 file represents a single sample or a small number of samples
If a single S3 file contains a lot of samples/rows, try the next submission strategy.
1
POST <batch_api_endpoint>:
2
{
3
    "workers": <int>,               # the number of workers to allocate for this job (required)
4
    "timeout": <int>,               # duration in seconds since the submission of a job before it is terminated (optional)
5
    "sqs_dead_letter_queue": {      # specify a queue to redirect failed batches (optional)
6
        "arn": <string>,            # arn of dead letter queue e.g. arn:aws:sqs:us-west-2:123456789:failed.fifo
7
        "max_receive_count": <int>  # number of a times a batch is allowed to be handled by a worker before it is considered to be failed and transferred to the dead letter queue (must be >= 1)
8
    },
9
    "file_path_lister": {
10
        "s3_paths": [<string>],     # can be S3 prefixes or complete S3 paths (required)
11
        "includes": [<string>],     # glob patterns (optional)
12
        "excludes": [<string>],     # glob patterns (optional)
13
        "batch_size": <int>,        # the number of S3 file paths per batch (the handle_batch() function is called once per batch) (required)
14
    }
15
    "config": {                     # arbitrary input for this specific job (optional)
16
        "string": <any>
17
    }
18
}
19
​
20
RESPONSE:
21
{
22
    "job_id": <string>,
23
    "api_name": <string>,
24
    "kind": "BatchAPI",
25
    "workers": <int>,
26
    "config": {<string>: <any>},
27
    "api_id": <string>,
28
    "sqs_url": <string>,
29
    "timeout": <int>,
30
    "sqs_dead_letter_queue": {
31
        "arn": <string>,
32
        "max_receive_count": <int>
33
    },
34
    "created_time": <string>
35
}
Copied!
The entire job specification is written to /cortex/spec/job.json in the API containers.
Newline delimited JSON files in S3
If your input dataset is a newline delimited json file in an S3 directory (or a list of them), you can define delimited_files in your request payload to break up the contents of the file into batches of size delimited_files.batch_size.
Upon receiving delimited_files, your Batch API will iterate through the delimited_files.s3_paths to generate the set of S3 files to process. You can use delimited_files.includes and delimited_files.excludes to filter out unwanted files. Each S3 file will be parsed as a newline delimited JSON file. Each line in the file should be a JSON object, which will be treated as a single sample. The S3 file will be broken down into batches of size delimited_files.batch_size and submitted to your workers. To learn more about fine-grained S3 file filtering see filtering files.
The total size of a batch must be less than 256 KiB.
This submission pattern is useful in the following scenarios:
one or more S3 files contains a large number of samples and must be broken down into batches
1
POST <batch_api_endpoint>:
2
{
3
    "workers": <int>,               # the number of workers to allocate for this job (required)
4
    "timeout": <int>,               # duration in seconds since the submission of a job before it is terminated (optional)
5
    "sqs_dead_letter_queue": {      # specify a queue to redirect failed batches (optional)
6
        "arn": <string>,            # arn of dead letter queue e.g. arn:aws:sqs:us-west-2:123456789:failed.fifo
7
        "max_receive_count": <int>  # number of a times a batch is allowed to be handled by a worker before it is considered to be failed and transferred to the dead letter queue (must be >= 1)
8
    },
9
    "delimited_files": {
10
        "s3_paths": [<string>],     # can be S3 prefixes or complete S3 paths (required)
11
        "includes": [<string>],     # glob patterns (optional)
12
        "excludes": [<string>],     # glob patterns (optional)
13
        "batch_size": <int>,        # the number of json objects per batch (the handle_batch() function is called once per batch) (required)
14
    }
15
    "config": {                     # arbitrary input for this specific job (optional)
16
        "string": <any>
17
    }
18
}
19
​
20
RESPONSE:
21
{
22
    "job_id": <string>,
23
    "api_name": <string>,
24
    "kind": "BatchAPI",
25
    "workers": <int>,
26
    "config": {<string>: <any>},
27
    "api_id": <string>,
28
    "sqs_url": <string>,
29
    "timeout": <int>,
30
    "sqs_dead_letter_queue": {
31
        "arn": <string>,
32
        "max_receive_count": <int>
33
    },
34
    "created_time": <string>
35
}
Copied!
The entire job specification is written to /cortex/spec/job.json in the API containers.
Get a job's status
1
cortex get <batch_api_name> <job_id>
Copied!
Or make a GET request to <batch_api_endpoint>?jobID=<jobID>:
1
GET <batch_api_endpoint>?jobID=<jobID>:
2
​
3
RESPONSE:
4
{
5
    "job_status": {
6
        "job_id": <string>,
7
        "api_name": <string>,
8
        "kind": "BatchAPI",
9
        "workers": <int>,
10
        "config": {<string>: <any>},
11
        "api_id": <string>,
12
        "sqs_url": <string>,
13
        "status": <string>,
14
        "batches_in_queue": <int>   # number of batches remaining in the queue
15
        "worker_counts": {          # worker counts are only available while a job is running
16
            "pending": <int>,       # number of workers that are waiting for compute resources to be provisioned
17
            "initializing": <int>,  # number of workers that are initializing
18
            "running": <int>,       # number of workers that are actively working on batches from the queue
19
            "succeeded": <int>,     # number of workers that have completed after verifying that the queue is empty
20
            "failed": <int>,        # number of workers that have failed
21
            "stalled": <int>,       # number of workers that have been stuck in pending for more than 10 minutes
22
        },
23
        "created_time": <string>
24
        "start_time": <string>
25
        "end_time": <string> (optional)
26
    },
27
    "endpoint": <string>
28
    "api_spec": {
29
        ...
30
    },
31
    "metrics": {
32
        "succeeded": <int>      # number of succeeded batches
33
        "failed": int           # number of failed attempts
34
        "avg_time_per_batch": <float> (optional)  # average time spent working on a batch (only considers successful attempts)
35
    }
36
}
Copied!
Stop a job
1
cortex delete <batch_api_name> <job_id>
Copied!
Or make a DELETE request to <batch_api_endpoint>?jobID=<jobID>:
1
DELETE <batch_api_endpoint>?jobID=<jobID>:
2
​
3
RESPONSE:
4
{"message":"stopped job <job_id>"}
Copied!
Additional Information
Filtering files
When submitting a job using delimited_files or file_path_lister, you can use s3_paths in conjunction with includes and excludes to precisely filter files.
The Batch API will iterate through each S3 path in s3_paths. If the S3 path is a prefix, it iterates through each file in that prefix. For each file, if includes is non-empty, it will discard the S3 path if the S3 file doesn't match any of the glob patterns provided in includes. After passing the includes filter (if specified), if the excludes is non-empty, it will discard the S3 path if the S3 files matches any of the glob patterns provided in excludes.
If you aren't sure which files will be processed in your request, specify the dryRun=true query parameter in the job submission request to see the target list.
Here are a few examples of filtering for a folder structure like this:
1
├── s3://bucket
2
    └── images
3
        ├── img_1.png
4
        ├── img_2.jpg
5
        ├── img_3.jpg
6
        └── img_4.gif
Copied!
Select all files
1
{
2
    "s3_paths": ["s3://bucket/images/"]
3
}
4
​
5
# or
6
​
7
{
8
    "s3_paths": ["s3://bucket/images/img"]
9
}
10
​
11
# Would select the following files:
12
# s3://bucket/images/img_1.png
13
# s3://bucket/images/img_2.jpg
14
# s3://bucket/images/img_3.jpg
15
# s3://bucket/images/img_4.gif
Copied!
Select specific files
1
{
2
    "s3_paths": [
3
        "s3://bucket/images/img_1.png",
4
        "s3://bucket/images/img_2.jpg"
5
    ]
6
}
7
​
8
# Would select the following files:
9
# s3://bucket/images/img_1.png
10
# s3://bucket/images/img_2.jpg
Copied!
Only select JPG files
1
{
2
    "s3_paths": ["s3://bucket/images/"],
3
    "includes": ["**.jpg"]
4
}
5
​
6
# Would select the following files:
7
# s3://bucket/images/img_2.jpg
8
# s3://bucket/images/img_3.jpg
Copied!
Select all JPG files except one specific JPG file
1
{
2
    "s3_paths": ["s3://bucket/images/"],
3
    "includes": ["**.jpg"],
4
    "excludes": ["**_3.jpg"]
5
}
6
​
7
# Would select the file:
8
# s3://bucket/images/img_2.jpg
Copied!
Select all files except GIFs
1
{
2
    "s3_paths": ["s3://bucket/images/"],
3
    "excludes": ["**.gif"]
4
}
5
​
6
# Would select the files:
7
# s3://bucket/images/img_1.png
8
# s3://bucket/images/img_2.jpg
9
# s3://bucket/images/img_3.jpg
Copied!
Containers
Statuses
Last modified 15d ago
Copy link
Contents
Get the Batch API's endpoint
Submit a Job
Data in the request
S3 file paths
Newline delimited JSON files in S3
Get a job's status
Stop a job
Additional Information
Filtering files