ecFlow looks for files using the following search process, when trying to locate the '.ecf' associated with a task.
- First it uses the variable ECF_SCRIPT and tries to open that file. ECF_SCRIPT is generated from ECF_HOME/SUITE/FAMILY/TASK.
- Otherwise if variable ECF_FILES exists, it must point to a directory which is searched in reverse order, e.g. let's assume that the node name is /o/12/fc/model and that ECF_FILES is defined as /home/ecmwf/emos_ECF/def/o/ECFfiles.
The order of files tried is as follows:
/home/ecmwf/emos_ECF/def/o/ECFfiles/o/12/fc/model.ecf
/home/ecmwf/emos_ECF/def/o/ECFfiles/12/fc/model.ecf
/home/ecmwf/emos_ECF/def/o/ECFfiles/fc/model.ecf
/home/ecmwf/emos_ECF/def/o/ECFfiles/model.ecf
If the directory associated with ECF_FILES does not exist. The server will try variable substitution, and repeat the backward search - Finally it searches ECF_HOME directory. ( in reverse order)
This may at first may be seen as over kill, but you can put all the files for a number of suites in one distinct file system/directory.
If the original ECF_SCRIPT did not exist, ecFlow will check the directories for the job file in ECF_HOME (ECF_SCRIPT is derived from ECF_HOME.). If a directory does not exist, ecFlow will create it. This helps to clean up old job-files and output and makes the maintenance of the scripts easier. It also guarantees that the output can be redirected into the file without the job creating the directory. (e.g. NQS option QSUB -ro, or when using redirection.)
Using ECF_FILES means that you do not have to create and maintain a link-jungle, e.g. the model.ecf above exists in a number of different families in ECMWF operational suites. The file is placed in a directory .../ECFfiles/fc/ and used by nodes /o/00/fc/model, /o/12/fc/model etc. This trick works nicely as long as there are no other tasks named model in the same family.