Present: Peter C, Frank G, Clara G, Stefan S, Alexandros K, Fernando V, Piotr G, Valeri F (over zoom), Georgios L (over zoom)
During a lively discussion during the presentation the following actions have been identified
- it would be useful to have a script for a massive-deactivation of license tickets (not only for activation)
- one should add a picture describing the architecture and processes
- while performing a test with a massive number of EvMan's (or projects) connecting to the license server, measure the CPU consumption
-> this would be meaningful to decide on an architecture with 2 servers, e.g. whether they should both be used at the same time (load sharing) or one should be a "master" and another a "warm standby" activated when the master becomes unavailable, etc - verify if the first failed license check is also reported via a datapoint
- when the 3rd license check fails, and the PARA is deactivated this does not allow to save the work; hence we should have a well visible popup/notification in GEDI (not necessarily in PARA) to inform the person of the work that is unsaved. The popup should be shown possibly after the first failed check, and e.g. change the background of GEDI, etc, etc.
-> maybe we could develop a script that does it ourselves, without involving ETM... - Check whether license-recovery is still attempted during the "7 day license mode"
-> if not, probably our script should do this - Check if we could tell to the license-switch to give us "any" license (ie. CodeMeter would pick one for us) that is available, rather than specifying exactly which one we want
- an interesting scenario for a "emergency mini license server", as an alternative to USB dongles: by default all the machines would have a local license, plus there will be a machine running a license server with "a few" licenses that could be used immediately when a machine is replaced (ie. with no need for online activation at Wibu servers).
-> ALICE would still prefer to have a bunch of physical dongles
-> however, to plug a dongle to a machine you still need to get to the counting room, and do it, which takes time and ain't that simple
-> typically, if a proxy server exists on the experiment network the activation process should be reliable and effortless (on small scale) - a central tool (add-on for the fwSystemOverview maybe?) that would allow to command the license activation/deactivation remotely, and deliver diagnostics
- we need to test the behaviour of the licensing mechanism in virtualized environment of LHCb. Will the license remain valid if the VM gets spawned on another host? Would recovery from a snapshot work? This may be very much dependent on the virtualization solution - so far it seems to work pretty OK with VirtualBox
- Backup of license containers: possibly verify if possible at all, and if one could, for instance, replace a failed disk and restore from backup
- LHCb and CMS interested in pursuing the tests with the license server; LHCb keen to try a "warm standby" setup;
- ALICE prefers local activation of licenses
- ATLAS will most likely stay with local licenses at least initially
- in any case, more tooling for centralized overview/maintenance and possibly the automatism necessary for license-switching are needed, whatever the architectural choice is in each of the experiments
There are minutes attached to this event.
Show them.