JCOP FWWG 12 July 2022
Present: Christophe Pires, Luis Miguens Fernandez, Piotr Golonka, Wassef Karimeh, Ombretta Pinazza, Piero Giorgio Verdini, Lukasz Goralczyk, Valeri Falaleev, Cristina Vazquez, Mariusz Suder, Mateusz Lechman, Stefan Schlenker, Geraldine Thomas, Clara Gaspar
Indico: https://indico.cern.ch/event/1169926/
FwDIP
Luis presented a set of slides summarizing the open issues for the fwDIP component. The issues have been discussed as they were presented - the notes below snapshot the discussion; each first-line bullet point correspond to a slide from the presentation (attached in INDICO)
- Invalidity of data when starting the DIP Manager (ENS-30651 from CMS)
- ATLAS: cmdline arg is slightly confusing (could be called differently, could be looked up in the doc). Current behaviour is not a problem
- CMS: the problem is with subscriptions for which you got data long time (days) ago; this would invalidate data, and it would impact data analysis
- The archive would anyway have two values - the one valid and the one invalid. Yet there are some weird alarms related to DIP publications becoming invalid , and the two entries with timestamp differs - reconstructing this could be problematic, but maybe you would like to know that you didn't get an update
- The problem is with the restart of the client...
- Illustration of problem for CO2 cooling in CMS pixels - representative temperature transmitted by DIP. If there is a stale value, not representative ? You should also subscribe to DIP-quality as it is set by the manager. The actual problem is opposite: the manager sometimes invalidates the data that is actually valid - this is at the startup. This affects offline analysis, hence no danger to the hardware... There would be no impact on the "online" behaviour
- for ATLAS: the combination of bullet 1 and 2 of the slide would be optimal.
- CMS: the drawback would be that a simple restart of the manager would trigger a callback... Why is it that DIP client decides on the quality? Shutting it down should have nothing to do with it... For CMS point 3 would be not to have it modified
- ATLAS would absolutely need to have the way of invalidating when DIP manager is not running - namely DIP is used to subscribe to DSS and one would need to know it. A possibility of having it in necessary.
- Currently it is at the startup that invalidation happens... Redoing this is a significant effort
- For CMS redu-switchovers could trigger a restart of managers hence should be smooth...
- We could also use other bits, not necessarily the invalid bit.
- there should be an option to have it as invalid (ATLAS) as it is e.g. in OPCUA,
- To be checked by Luis: what happens at redu switchover (it means killing DIP managers). During a switchover one should not get a invalid bit. But this is not a problem of the switchover.
- Proposal: a flag in the config file
- TO FOLLOW UP: check the bit 3 behaviour - ie. the one that should indicate "uncertainty
- Default behaviour when receiving invalid data or any unhandled exception (ENS-30413)
- proposal: setting the invalid bit immediately...
- "corrupt"? what does this mean? Actually this happens when DIM-bridge was used - its publishers under some conditions got a wrong/unformat format... Hence actually was happening... But this problem at the DIM layer is fixed at the root cause.
- so far there was no way to react to it as DIP API did not call back in such case - only logged a message. Now we have it...
- CMS: this should be correct - data not written to a DP (invalidated) - again, it is not that the publisher invalidated it...
- Maybe requesting an update of a value would be sensible - but what if we get a corrupted data again? We should mark it somehow... Maybe another user bit?
- ATLAS: we would like to see it in a datapoint somehow! If it could be recovered, then fine, but certainly not ignore it
- Maybe one should again use the concept of the "unknown"?
- Agreeda config file option to determine how to signal it... ie. whether to propagate it to the invalid bit; default is to set the invalid bit (and a user bit)
- Improve the robustness when a DP does not exist or DP is deleted during run time...
- Hence, still make it possible that managers start
- the published value will get invalid instead of DIP manager crashing
- CMS: should the publication disappear? Then the client would learn about in anyway...
- CMS: the fact that DIP does not start is a clear message to the expert actually
- The crash on the DP actually restored the situation... but we should not rely on it ; abort
- But this is not good for UNICOS engineering... Restarting the manager
- ACTION: REVIEW WITH THE UNICOS TEAM
- UI improvements - importing/exporting configurations
- CMS: in some cases one may want to keep the system names... to publish the values for remote systems.
- It was already made "intelligent" to prefix the system name for non-local DPs.
- The problems would be only for OLD configurations (2017...) CMS has everything migrated by now...
- FWDIP-80 could be closed?
- NO! CMS cannot have it closed. This is related to storing the information in the ConfigurationDB. THERE WAS A MISINTERPRETATION ABOVE. UNDERSTOOD NOW. It is about how the DPs are stored in DP not in XML!
- others to be kept
- NA62 uses the UI to configure the publications, and it would be very welcome to have it improved.
CMS is using it too - often for tests.
- Custom user bits
- customize user bits to match the quality string of the DIP
- ENS-21856 - a case from Will Booth, originating from COOL for ATLAS. May need to be revisited
- Duplication of DIP "messages" (packets)
- CMS experienced this issue with the LHC handshake (STANDBY and OK with the very same timestamp, and hence STANDBY was missed)...
- Not easily reproducible. But should be fixed.
- The published should fix it in the first case! DIP should pass it over
- Clara: the publisher should pass both of them! CMS may also wants it
- Log it (that you received the publication twice), so it is not unnoticed. It will be set by WinCC OA to have a invalid bit due to duplicated timestamp.
- DIP logging
- ANY OTHER SPECIFIC ISSUE
- CMS: FWDIP-80: WAS MISUNDERSTOOD BEFORE; clarified the intent. Cannot be closed.
- CMS: DIP with same timestamp but different status - checking the content/quality... If data is different then instead of WinCC OA setting invalid because of same timestamp, DIP manager could increase the timestamp for the second value
- commandline options vs config file
- do not get rid of them completely - may still be used for testing
- keep both, hence assure backward compatible
- PlatformDependent.lib ... to be replaced with boost or stdc++
- we need to do it at some moment as it becomes unmaintanted
- it is used by fwDIP, not the DIP API, that is why the new maintainers are abandoning it; we are the only users
- How about using Qt? Could be an option
- As the startup sequence of manager would need to be refactored , QFuture would be very elegant
- it is also interconnected with logging
- Looks like a consolidation of the complete component
- Some code (SString) could be brought from PlatformDependent to fwDIP... But constructs got deprecated...
- Log4CPlus
- nightmare to get the log4c+ to work when compiling/debugging...
- The DIP API compiled/linked statically, hence the sources could be included somehow... A pipeline?
- to be seen.
AOB
- Good progress with the MVVM UI design pattern during the "Critical Mass" spring at BE-ICS. Very promising and well based starting point for fwTrending redesign. Will be presented at the JCOP FWWG in September. We plan a "minimum viable product" release of redesigned fwTrending to become available in ~3 weeks, installable and testable
- FwCore-8.5.0 will be available quite soon, featuring the new FwException CTRL++ class in fwGeneral subcomponent, as well as MODBUS addressing support for fwDevice (contributed by CMS) and fixes/enhancements for MQTT addressing in fwConfigs. Should arrive before the end of July
- WinCC OA 3.19 - very interesting features ahead - presentation attached.
- Next FWWG: after the Summer break - in September
There are minutes attached to this event.
Show them.