Download
operational requirements for core services n.
Skip this Video
Loading SlideShow in 5 Seconds..
“ Operational Requirements for Core Services ” PowerPoint Presentation
Download Presentation
“ Operational Requirements for Core Services ”

“ Operational Requirements for Core Services ”

155 Vues Download Presentation
Télécharger la présentation

“ Operational Requirements for Core Services ”

- - - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - - -
Presentation Transcript

  1. “Operational Requirements for Core Services” James Casey, IT-GD, CERN CERN, 21st June 2005 SC4 Workshop

  2. Summary • Issues as expressed by sites • ASGC, CNAF, FNAL, GRIDKA, PIC, RAL, TRIUMF • My synopsis of the most important issues • Where we are on them… • What are possible solutions in longer term CERN IT-GD

  3. ASGC - Features missing in core services • Local/remote diagnostic tests to verify the functionality and configuration. • This will be helpful for • Verifying your configuration • Generating test results that can be used as the basis for local monitoring • Detailed step-by-step troubleshooting guides • Example configurations for complex services • e.g VOMS, FTS • Some error message can be improved to provide more information to facilitate troubleshooting CERN IT-GD

  4. CNAF - Outstanding issues (1/2) • Accounting (monthly reports): • CPU usage in KSI2K-days  DGAS • Wall-clock time in KSI2K-days  DGAS • Disk space used in TB • Disk space allocated in TB • Tape space used in TB • Validation of raw data gathered, by comparison via different tools • Monitoring of data transfer: GridView and SAM? • More FTS monitoring tools necessary • (traffic load per channel, per VO) • Routing in LHC Optical Private Network? • Backup connection to FZK becoming urgent, and a lot of traffic using the production network infrastructure, between non-associated T1-T1 and T1-T2 sites CERN IT-GD

  5. CNAF – Outstanding Issues (2/2) • Implementation of a LHC OPN monitoring infrastructure still in its infancy • SE Reliability when in unattended mode: greatly improved with latest Castor2 upgrade • Castor2 performance during concurent import and export activities CERN IT-GD

  6. FNAL – Middleware additions • It would be useful to have better hooks in the grid services to enable monitoring for 24/7 systems • We are implementing our own tests to connect to the paging system • If the services had reasonable health monitors we could connect to it might spare us re-implementing or missing an important element to monitor CERN IT-GD

  7. GRIDKA – Feature Requests • improved (internal) monitoring • developers not always seem to be aware that hosts can have more than 1 network interface. • It should be that hosts can be reached via their long living alias and the actual name is unimportant (for reachability, not for security). • Error messages should make sense and be human readable! • simple example : • $ glite-gridftp-ls gsiftp://f01-015-105-r.gridka.de/pnfs/gridka.de/ • (typo in the hostname ^^^) • t3076401696:p17226: Fatal error: [Thread System] GLOBUSTHREAD: pthread_mutex_destroy() failed • [Thread System] mutex is locked (EBUSY)Aborted CERN IT-GD

  8. PIC – Some missing Features • All in general: • Clearer error messages • Difficult to operate (eg, it should be possible to reboot a host without affecting the service) • SEs: • Missing a procedure for “draining” an SE or gently “take it out of production” • Difficult to control access: for some features to be tested need the SE published in the BDII, but once is there there is no way to control who can access • Glite-CE: • A simple way to gather the DN of the submitter, having the Local Batch jobID (GGUS-9323) • FTS: • Unable to delete a channel which has “cancelled” transfers • Difficult to see a) that the service is having problems, and b) then to debug them CERN IT-GD

  9. RAL – Missing Features in File Transfer Service • Could collect more information (endpoints) dynamically • This is happening now in 1.5 • Logs • Comparing a successful and failed transfer is quite tricky • I can show you two 25 line logs, one for a failed and one for a successful srmcopy. The logs are completely identical. • Having logs files that are easy to parse for alerts or errors is of course very useful. • Offsite monitoring • How do we know a service at CERN is dead? • And what is provided to interface it to local T1 monitoring. CERN IT-GD

  10. TRIUMF – Core Services (1/2) • 'yaim', like any tool that wants to be general and complete, ends up being complicated to implement, to debug and to maintain. • In trying to do a lot from two scripts (install_node and configure_node) and one environment file (node-info.def) bypasses some basic principles of unix system management: • use small, independent tools, and combine them to achieve your goal. • Often a 'configure_node' process needs to be run multiple times to get it right. • It would help a lot if it did not repeat useless, already completed, time-consuming 'config_crl'. CERN IT-GD

  11. TRIUMF – Core Services (2/2) • An enhancement for the yaim configure process: • it would also be useful if the configure_node process would contain a hook to run a user-defined post-configuration step. • There is frequently some local issue that needs to be addressed, and we would like to have a line in the script that calls a local, generic script that we could manage, and would not be over-written during 'yaim' updates. • The really big hurdle will always be Tier 2's (large number of sites out there). • The whole process is just difficult for the Tier 2's. • It doesn't really matter all that much what the Tier 1's say - they will andmust cope. • One should be aggressively soliciting feedback from the Tier 2's. CERN IT-GD

  12. Top 5…. • Better logging • Missing Information (e.g. DN in transfer log) • Hard to understand logs • Better diagnostics tools • How do I verify my configuration is correct? • … and functional for all VOs? • Toubleshooting guides • Better error messages from tools • Monitoring • … and interfaces to allow central/remote components to be interfaced to local monitoring system CERN IT-GD

  13. Logging • FTS Logs have several problems: • Only access to logs via interactive login on transfer node • Plans to have full info in DB • Will come after schema upgrade in next FTS release • CLI tools/web interface to retrieve them • Intermediate stage is to have final reason in DB • Outstanding bug sets this to AGENT_ERROR for 90% of messages • Should be fixed soon (I hope!) • Logs not understandable • When SRM v2.2 rewrite is done, a lot of cleanup will (need to) happen CERN IT-GD

  14. Diagnostic tools/ Troubleshooting guides • SAM (Site Availability Monitoring) is the solution for diagnostics • Can run validation tests as any VO, and see the results • System is in infancy • Tests need expanding • But the system is very easy to write tests for • … and the web interface is quite nice to use • Troubleshooting guides • These are acknowledged needed for all services • T-2 tutorials helped in gathering some of these materials • Look at tutorials from last week in Indico for more info CERN IT-GD

  15. SAM 2 • Tests run as operations VO: ops • sensor test submission available for all VOs • critical test set for VOs (defined using FCR) • Availability Monitoring • aggregation of results over a certain time • site services: CE, SE, sBDII, SRM • central services: FTS, LFC, RB • status calculated in every hour → availability • current (last 24 hours), daily, weekly, monthly CERN IT-GD

  16. SAM Portal -- main CERN IT-GD

  17. SAM -- sensor page CERN IT-GD

  18. Monitoring • It’s acknowledged the GRIDVIEW is not enough • It’s good for “static” displays, but not good for interactive debugging • We’re looking at other tools to parse the data • SLAC have interesting tools for monitoring netflow data • This is very similar in format to the info we have in globus XFERLOGs • And they even are thinking of alarm systems • I’m interested to know what types of features such a debugging/monitoring system should have • We’d keep it all integrated in a GRIDVIEW like-system CERN IT-GD

  19. Netflow et. al. • Peaks at known capacities and RTTs • RTTs might suggest windows not optimized CERN IT-GD

  20. Mining data for sites CERN IT-GD

  21. Diurnal behavior CERN IT-GD

  22. One month for one site CERN IT-GD

  23. Effect of multiple streams • Dilemma what do you recommend: • Maximize throughput but unfair, pushes other flows aside • Use another TCP stack, e.g. BIC-TCP, H-TCP etc. CERN IT-GD

  24. Thank you … CERN IT-GD