Better handle failed processing jobs when terminated by cluster

Raised by Mark; this collection has a permanent spinning cog: https://ispyb.diamond.ac.uk/dc/visit/mx37593-40/id/18159036

This corresponds to a reprocessing run that combined several datasets:

The log file shows it terminating in the middle of integration (I assume due to OOM errors):

...
oooooooooooooooooooooooooooooooooo.ooooooooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo
oooooooooooooooooooooooooooooooooooooooooooooooooooooooooooo
"o" => good        "%" => ok        "!" => bad rmsd
"O" => overloaded  "#" => many bad  "." => weak
"@" => abandoned
Mosaic spread: 0.413 < 0.413 < 0.413
-------------------- Spotfinding SWEEP2 --------------------
67030 spots found on 3600 images (max 1798 / bin)
              ***
            ******
           *********                       ****
         *************                  *********
*     ******************* ***          *************     ***
** *****************************   ******************** ****
************************************************************
************************************************************
************************************************************
************************************************************
1                         image                         3600
------------------- Autoindexing SWEEP2 --------------------
All possible indexing solutions:
aP  40.40  57.05  77.35 102.77  94.67 100.11
Indexing solution:
aP  40.40  57.05  77.35 102.77  94.67 100.11
-------------------- Integrating SWEEP2 --------------------

possibly we need to have a service that occasionally checks the cluster status for known reprocessing jobs, and marks it as failed if the cluster job has gone away but no status update written?

Nenhum arquivo, teste ou ponto de entrada é nomeado. Comece rastreando como os jobs de reprocessamento recebem atualizações de status e como o encerramento do cluster é representado; está concluído quando um job conhecido é marcado como falho ao desaparecer o job do cluster sem uma atualização de status, em vez de permanecer em um estado de carregamento contínuo.