FWWG

Wednesday 21 Nov 2007, 14:00 → 16:00 Europe/Zurich

13/3-005

Slides Presentation_21st_November_2007.pdf Presentation_21st_November_2007.pptx

Participants: Piotr, Francisca, Oliver, Manuel, Peter, Lennart, Clara, Niko, Jim, Stefan, Fernando, Bobby, Frank, Wayne 1) Software and application installation, maintenance, backup and restore strategies Wayne presented a summary of the approach of each of the experiments to software and application installation, maintenance, backup and restore. His slides can be found under: http://indico.cern.ch/conferenceDisplay.py?confId=24225 There were a number of discussions points/clarifications and questions as follows: ALICE: - It is not possible to copy a project while it is running as there is a risk of leaving the PVSS database in an inconsistent state which may prevent the project from starting. - It was requested that a study of the on-line B/U be performed to understand the impacts on a running project, i.e. how long does it take in a typical representative project, what does it actually copy, etc. - Bobby raised the point that if there are project paths in the config file which are not available then a project will not start - Peter stated that they have versioning based on date - The script which copies panels and scripts runs periodically but only copies file sif a change is detected - CMF is currently not used to install OPC servers - ALICE has hot spare machines which have the OS installed but these are not in the domain - The FS has RAID disks and a copy is also made to another disk ATLAS: - ATLAS create a dplist and FSM package for projects because they have seen occasions where a running project is corrupted and installing the B/U does not cure the problem. Hence, they need to be able to rebuild the project - Stefan was asked which OPC servers were already packaged. He agreed to send a list to Wayne - Frank asked whether the tool to check the parameterization history was available. Manual will provide this - Stefan noted that the ATLAS controls PC have RAID disk and this reduces the risk of losing a project due to a disk failure CMS: - When fw components are installed, files are not overwritten if they already exist - CMS currently intends to keep two versions of the fw components but as a complete package and not individually, i.e. one can installed CMS fw package n or revert back to CMS fw package n-1 - It is clear that although the central team forces installation of new versions of fw components or PVSS patches, this has to be coordinated with the sub-detector teams - Piotr noted that PVSS is packaged as PRMs for Linux and these can be installed automatically using LinuxFC - As projects rely on the FS what happens if it is not available? Basically the risk is low, as there is considerable redundancy built in, and if it is not available then operation will be degraded until it is available again - CMS have hot spares with OS and all basic S/W, including PVSS with license (based only on MAC address), already installed - For OPC servers not using configuration files, CMS has a script with automatically configures the OPC server (writing to the Windows registry). The other experiments would be interested in this script - PVSS projects are created automatically using a script and based on information in the config DB LHCb: - It seems that SMS offers all the facilities that LHCb needs - The sub-detectors need to configure the OPC servers which means there will be one sub-detector account with local administration privileges. This policy may be changed in the longer-term - The PC will all be installed with the same set of software even if some is not required, e.g. CAEN OPC server for a machine with no connection to CAEN. This is felt to be easier to manage than many different configurations - LHCb will license the shareware product Service Plus for managing programs that should run automatically. Other experiments expressed an interest in this tool - CMS reported an issue with the DIM DNS hanging requiring regular restarts - Clara reported an issue whereby if the config file has many project path entries it takes a very long time for the UIM to start. This has been reported to ETM. - Due to the need to be able to kill hanging OPC servers by people without the necessary level of privilege, LHCb has created a Task Manager which runs with administrator privileges - It was agreed that IT/CO would investigate the possible B/U possibilities and make recommendations 2) Patch Release Mechanism Manuel asked whether the PVSS Patch Release strategy applied by IT/CO was okay. It was agreed that only cumulative patches need to be made available. Individual patches can be removed. However, the RDB patches should be handled separately, i.e. not included in the cumulative patch and hence provided individually. A fw component will be provided to perform manual steps in the installation procedure to allow the installation to proceed automatically from a user point of view. 3) CAEN OPC Server/PVSS OPC Client Issue Bobby reported an issue with the CAEN OPC server and PVSS. The CAEN OPC server determines what equipment is connected at start-up and builds the appropriate name space. If equipment which is defined in PVSS is not accessible the PVSS OPC client sees this and creates a log message. However, it does not set the invalid bit and from a user point of view he says the data available as if everything was fine, but potentially with a random value which is not valid. 4) PVSS 3.7 Manuel reported that IT/CO was testing 3.7 but with a low priority. He encouraged the experiments to also test it as they might discover problems not seen by IT/CO. The experiments have been asked to consider whether they will migrate to 3.7 or not, and if so, when. 5) RDB Further Work Manual gave a brief summary of the work performed during the week in which Ewald Sperrer from ETM was at CERN. Good progress was made in fixing some bugs identified by ATLAS and with the design for the new requirements. It was agreed that ETM would work on the C++ code in the RDB Manager and CERN on the PLSQL code for the database. Two further periods have been schedule for working together; one in December and the other in January. Large scale tests have also been scheduled for January.