* http://localhost:8080/playground - A GraphQL playground. To use it you must have a authenticated session in the same browser.
* http://localhost:8080/swagger - A Swagger UI. To use it you have to be logged out, so no user session in the same browser. Use the JWT token with role Api generate previously to authenticate via http header.
### Use cc-backend API to start job
* Enter the URL `http://localhost:8080/swagger/index.html` in your browser.
* Enter your JWT token you generated for the API user by clicking the green Authorize button in the upper right part of the window.
* Click the `/job/start_job` endpoint and click the Try it out button.
* Enter the following json into the request body text area and fill in a recent start timestamp by executing `date +%s`.:
```
{
"jobId": 100000,
"arrayJobId": 0,
"user": "ccdemouser",
"subCluster": "main",
"cluster": "emmy",
"startTime": <date+%s>,
"project": "ccdemoproject",
"resources": [
{"hostname": "e0601"},
{"hostname": "e0823"},
{"hostname": "e0337"},
{"hostname": "e1111"}],
"numNodes": 4,
"numHwthreads": 80,
"walltime": 86400
}
```
* The response body should be the database id of the started job, for example:
```
{
"id": 3937
}
```
* Check in ClusterCockpit
- User `ccdemouser` should appear in Users-Tab with one running job
- It could take up to 5 Minutes until the Job is displayed with some current data (5 Min Short-Job Filter)
- Job then is marked with a green `running` tag
- Metricdata displayed is read from cc-metric-store!
### Use cc-backend API to stop job
* Enter the URL `http://localhost:8080/swagger/index.html` in your browser.
* Enter your JWT token you generated for the API user by clicking the green Authorize button in the upper right part of the window.
* Click the `/job/stop_job/{id}` endpoint and click the Try it out button.
* Enter the database id at id that was returned by `start_job` and copy the following into the request body. Replace the timestamp with a recent one:
```
{
"cluster": "emmy",
"jobState": "completed",
"stopTime": <RECENTTS>
}
```
* On success a json document with the job meta data is returned.
* Check in ClusterCockpit
- User `ccdemouser` should appear in Users-Tab with one completed job
- Job is no longer marked with a green `running` tag -> Completed!
- Metricdata displayed is now read from job-archive!
* Check in job-archive
-`cd ./cc-backend/var/job-archive/emmy/100/000`
-`cd $STARTTIME`
- Inspect `meta.json` and `data.json`
## Helper scripts
* In this tarball you can find the perl script `generate_subcluster.pl` that helps to generate the subcluster section for your system.
Usage:
* Log into an exclusive cluster node.
* The LIKWID tools likwid-topology and likwid-bench must be in the PATH!
*`$./generate_subcluster.pl` outputs the subcluster section on `stdout`
Please be aware that
* You have to enter the name and node list for the subCluster manually.
* GPU detection only works if LIKWID was build with Cuda avalable and you run likwid-topology also with Cuda loaded.
* Do not blindly trust the measured peakflops values.
* Because the script blindly relies on the CSV format output by likwid-topology this is a fragile undertaking!
