JCOP FWWG Meeting 4 February 2020
New Alarm Screen: update on the progress, demo and technicalities
- News on JCOP Fw and WinCC OA
- WinCC OA 3.16 Patch P016 released for CERN users;
- no fix for lm_ip problem yet - we will wait for P017 as it may bring more fixes
- New patch for WinCC OA 3.15 announced by ETM
- new patches for WinCC OA 3.15 are not interesting/needed anymore for the experiments, as everyone update to WinCC OA 3.16 now; we may repackage it but no need for validation
- Projects startup/shutdown on Linux: issue with projects not being stopped correctly at machine shutdown. A proper solution is to rework it to use systemd, rather than use a wrapper. A draft of document describing this is in internal review and will be circulated shortly. The document will describe how to run projects as OS service and how to configure this through systemd.
- Rafal's presentation of the new Alarm Screen
- Suggestions related to distributed system selection/monitoring/reporting
- The selector should contain "All" and "None" entries - important for large systems
- The list of systems that should be connected should be explicitly defined in the configuration, as it is the case in the current alarm screen. Simply getting this from the config file of the project is not sufficient. [addition PG: think of config files for scattered-UIs - they do not contain distPeer definitions; think of the example of dist connection from NA62DCS to central monitoring system (MOON) for which alarms should be explicitly excluded from the alarm screen]
- There should be a very visible indication that not all systems are connected
- The dist-system selector button should be protected by access control - shifters should not be allowed to change it. Should it be hidden completely for them?
- Linked to the two above: still there should be a way of getting the detailed information about which systems out of desired ones are connected [PG addition: overlap of functionality with unDistributedControl]
- there are particular cases such as system restart or redu-switch in which we would like to reduce "flickering" of the alarm screen: the alarms should be "cached for some seconds" (parameterizable). This is a new request - the functionality is not available in the current alarm screen
- Rafal clarified that in addtition to tracing the distconnection datapoints the new sysConnect() method is using to trace system connect/disconnect.
- Use of stylesheets:
- in Rafal's demo they are applied with the "-stylesheet" commandline option;
- clarification from Piotr: there is a background project in BE-ICS to explore the potential of stylesheets, in particular to provide a "dark style"; activity on this will be reported
- Special columns
- in the current fwAlarmScreen there is a custom column that shows comment-or-alias-or-dpeName (ie falling back if one is empty to the next one). This is very much used in LHCb. Double check that it is available in the new AS.
- Customizable right-click menu
- it is already available, with content and callback interfaces via CTRL; LHCb needs this for sending an alarm to their log book
- Performance
- Is the new alarm screen single threaded? So far it is sufficient even for 100.000 of alarms filtered "live". Test with generating an avalanche or constant flow of alarm updates were performed; the new alarm screen no problem with any "reasonanble" rate of alarms. Multi-threading may become complex in the parts that needs to refer to WinCC OA API (which is not thread-safe). If need, we could consider it.
- Use case in CMS during shutdowns: 60-70k alarms permanently on the screen and people need to watch/filter on them. For this the indication of how many "new" alarms there is mandatory (check/implement!) -> in the current fwAlarmScreen this is the functionality of "alarm group" boxes that indicate the number of new alarms.
- How about memory consumption? It depends on the configured columns. Alarms are stored only once and views refer to parts of this data in a very lightweight way... Would be interesting to see the number compared to the current alarm screen, for some of cases, such as an "empty" alarm screen, one with a handful of alarms, or large number. These tests are already envisioned in the test catalogue - will be completed with numbers before the next release.
- Multiple alarm screens in the same UI?
- typically there is one central alarm screen on a dedicated console. How then would one have any profit from the alarm-connect to all alarms?
- consider the case of a FSM Tree displaying the panel for a sector of a subdetector and showing a mini-list of alarms only for this sector; open a few of these for various sectors. In UNICOS this is the case of "device faceplates"
- Event screen
- use case: the "historical mode" used to analyze problems based on archived data
- Live ES could be interesting for LHCb, but would require more understanding of what is potentially possible.
- FwAlarmScreen deprecation?
- Certainly not before the end or Run3 for sure; will be kept (with all its limitations) as long as required. New alarm screen with production-grade quality certainly in 2020, on time to start data-taking.
- How much of functionality of the requirements was implemented?
- Some of critical features (such as masking, alarm reduction for ATLAS) must to be mandatorily available before any deployment in the control rooms.
- Piotr assured that after the upcoming release of the "preview version" all the experiments will be contacted to provide feedback, and implementation of the top outstanding necessary issues for the production deployment will be scheduled with priority (should be covered by the JCOP Plan of Work for 2002).
- AOB
- Proposal for the JCOP FWWG meeting/seminar/technical session about the NextGenArchiver some time in Spring. This could be done once we have the initial CERN release of NGA (soon). A presentation may then be arranged for in April. The format for it (JCOP FWWG meeting or a technical seminar/workshop) needs to be agreed based on interest and needs.
There are minutes attached to this event.
Show them.