Transcript Document
WLCG Monitoring Consolidation
NEC`2013, Varna Julia Andreeva CERN IT-SDC
Outline
Motivation and goals Organization of work Progress so far JINR contribution Conclusions NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 2
Motivation
Substantial part of manpower involved in the monitoring of the OSG/WLCG infrastructure was covered by EGI/InSPIRE funding which is coming to its’ end => considerable reduction of manpower The system put in place for monitoring of the EGI infrastructure was designed for a fully distributed operational model with strong contributions from the regional teams. While the WLCG operational model is more centralized. Re-scoping of the monitoring system to the WLCG infrastructure creates an opportunity for simplification The WLCG operational model implies major contribution from the LHC experiments. Therefore the WLCG monitoring system should be flexible enough to be easily adapted to the needs of the LHC experiments and should be inline with their operational modes NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 3
Goal of the project
The goal of the project is the consolidation of the WLCG monitoring This includes critical analysis of what is monitored, technology used deployment and support model This should allow to: reduce the complexity of the system ensure simplified and more effective operations, support and service management encourage an efficient deployment strategy, with a common development process unify, where possible, the implementation of the monitoring components Wherever reasonable, the effort should be aligned with the activities of the Agile Infrastructure Monitoring team at CERN.
As a result by autumn 2014 the monitoring services should be operated and modified (when required) with at least twice smaller team, compared to spring 2013 13-Sep-2013 4 NEC`2013, Varna, Julia Andreeva, CERN
Organization of work (1)
The project is split in two stages: First one which should be accomplished by the end of September: Review of the current systems and metrics used for WLCG monitoring Collecting and summarizing requirements from the LHC computing community to understand the changes needed in monitoring.
Highlight areas that are expected to be problematic Propose a revised strategy and architecture for WLCG monitoring Suggest implementation technologies and approaches for the transitions 13-Sep-2013 5 NEC`2013, Varna, Julia Andreeva, CERN
Organization of work (2)
Second stage: implementation of the new monitoring framework and components needed for the transition to the new toolchain. A first prototype should be ready by the end of 2013. All monitoring should have transitioned to the new approach by summer 2014.
Working group led by Pablo Saiz (CERN IT-SDC-MI) consists of representatives of
WLCG monitoring development teams (SAM and Dashboard) Agile monitoring infrastructure development team 4 LHC experiments WLCG operations team Sites are represented by the members of the WLCG operations team
Small taskforces on dedicated subjects Mailing list: [email protected]
Fortnightly meetings with summary reports http://go.cern.ch/6XQQ 13-Sep-2013 NEC`2013, Varna, Julia Andreeva, CERN 6
Review of the current systems used for WLCG monitoring Monitoring applications currently provided by the WLCG monitoring team can be split in 4 groups: Monitoring and accounting of the job processing activity Monitoring and accounting of data access and data transfers Remote testing and monitoring of the distributed infrastructure Tools for dissemination purposes Both experiments and sites were asked to provide input regarding usefulness of the existing applications, possible improvements and missing functionality (if any). Input from sites was provided through the WLCG operations team which sent a questionaire to the sites. The review did not include VO-specific monitoring systems developed inside the experiments According to the collected input, SAM and most of the Dashboard applications are considered to be useful or essential by the experiments ( at least ATLAS and CMS) and by the WLCG operations team. No important gaps in the monitoring functionality were identified.
Detailed list with applications and their usage, architecture and technology review are included in the backup slides of this presentation. 13-Sep-2013 7 NEC`2013, Varna, Julia Andreeva, CERN
How to move forward
Stop support of applications not used by the community Reduce scope Unify implementation Evaluate and apply new technologies NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 8
Reduce scope
Re-scoping to WLCG will allow to: reduce the complexity of our services - concentrate on things required by experiments (for example no need for OPS tests, availability reports are based on the experiment tests) - decrease number of service instances we have to support - pass responsibility for SAM central services used for the EGI infrastructure to Greece/Croatia To be followed up next week at the EGI Technical Forum 13-Sep-2013 NEC`2013, Varna, Julia Andreeva, CERN 9
Unifying implementations (Infrastructure testing and monitoring as an example)
NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 10
Infrastructure testing and monitoring
Most of effort of the WLCG monitoring team is currently dedicated to the infrastructure testing and monitoring.
Historically there are multiple implementations. This group of applications includes most of applications and services we currently support This area was identified as the main one where we can save most of effort and do things in a more efficient way 13-Sep-2013 11 NEC`2013, Varna, Julia Andreeva, CERN
Where we are now
Experiment defined metrics Nagios 3 different colours 3 different systems Experiment specific processing Nagios SAM probes SSB Message queue MRS POEM SSB WEB service ACE MyWLCG WEB service SUM ATP Hammer Cloud jobs Hammer Cloud server Hammer Cloud DB NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 WEB interface 12
Drawbacks of the current state
Variety of schemas for metric repositories Variety of processing engines Different implementation of data collectors Different implementation of APIs and visualization layers Not all components provide enough flexibility for introducing new metrics, for combining metrics , for integrating custom processing algorithms While functional testing (Nagios) and stress testing (Hammer Cloud) by its nature might dictate different implementation of the test submission mechanism, the following part of the chain looks very similar : collecting metrics, processing them (with generic or custom algorithm), providing access to data through APIs and UIs
Ideally the common workflow should be applied to any kind of monitoring data
13-Sep-2013 13 NEC`2013, Varna, Julia Andreeva, CERN
Should follow the real experiment workflows
Does our testing and monitoring check things which would allow to provide a realistic view of whether infrastructure can be effectively used by the experiments?
SAM tests are submitted through WMS which LHC experiments do not use (or do not intend to use in the close future) for their workflows SAM tests are pretty static and in general check only basic functionality of the services Not all experiments intend to use SAM/Nagios in future (ALICE).
One of the possibilities would be to perform tests inside the experiment pilots All experiments would like to be able to provide their own metrics to be taken into account for evaluation of site usability and performance NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 14
Site Status Board (SSB) as a generic solution for a metric store
SSB looks to be generic enough to be used as a common metric store.
Provides access to the snapshot and historical data Provides data compaction and archiving Has a built in mechanism for calculating status of a given instance over a given time range (similar to availability calculation) It has necessary concepts for combining metrics which can be of interest for a particular task (SSB views) Was successfully used for operations by ATLAS and CMS Having a common implementation for metric store would simplify a lot the current architecture allow to easily evaluate and apply new technologies (for data storage and processing) 13-Sep-2013 15 NEC`2013, Varna, Julia Andreeva, CERN
SAM Nagios tests
Where we would like to get
External processing engine Hammer Cloud Metrics defined by experiments Pledges T R A N S P O R T SSB-like metric store WEB service Built-in processing engine Downtime UI WEB service Topology description WEB service Profile description NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 16
Current status
Prototype is being deployed: SSB instance deployed Is being filled with SAM production test results Working on aggregation (availability calculation) SSB web interface ready: Summary page, current status, historical plots, ranking plots, ‘readiness’ plots 13-Sep-2013 17 NEC`2013, Varna, Julia Andreeva, CERN
And current metrics already published
NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 18
Evaluation of the new technologies
Following recommendations of the AI monitoring team, Elasticsearch was evaluated for several WLCG monitoring applications including SSB NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 19
WLCG Transfer Dashboard prototype using Elasticsearch as a storage solution
NEC`2013, Varna, Julia Andreeva, CERN For details on other layers, see: http://cern.ch/go/pw7 F See backup slide # for conclusions of current evaluation 13-Sep-2013 20
Packaging and deployment
Using openstack , puppet, hiera, foreman Quota of 100 nodes, 240 cores Multiple templates already created Development machine (9 nodes) Web servers ( SSB, SUM, WLCG transfers, Job: 20 nodes ) Elastic Search (6 nodes), Hadoop (4 nodes) Currently working on nagios installation Migrating machines from quattor to AI Koji and Bamboo integration for build system and continuous NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 21
JINR contribution
Russian institutions, in particular JINR are actively participating in the WLCG monitoring work including WLCG monitoring consolidation project Part of this work is performed in the framework of CERN IT Russia collaboration program (WLCG monitoring task) JINR team: Vladimir Korenkov (leading activity from the Russian side), Sergey Belov, Ivan Kadochnikov, Sergey Mitsin, Elena Tikhonenko, students from Dubna university JINR colleagues contribute to the development of the WLCG transfer Dashboard, xrootd Dashboards for ATLAS and CMS, evaluation of the NoSQL technologies for storage and processing of the WLCG monitoring data NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 22
Conclusions
First phase of the project consisting of the assessment of the current monitoring systems is accomplished A revised strategy and architecture for WLCG monitoring is being defined Working in a very close collaboration with experiments, operations, AI monitoring By spring 2014 the transition to a new toolchain should be performed. As a result by autumn 2014 the monitoring services should be operated and modified (when required) with at least twice smaller team, compared to spring 2013 13-Sep-2013 23 NEC`2013, Varna, Julia Andreeva, CERN
Backup slides
NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 24
Overview of the current monitoring applications (usage) (1)
NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 25
Overview of the current monitoring applications (usage) (2)
NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 26
Review of the current systems used for WLCG monitoring (architecture)
Regarding architecture, the Dashboard applications are all very similar and are developed in the common framework. For the SAM system, one of the main limitations reported in the review was the fact that various components of the system not always communicate through the APIs , but have shared tables in the DB, which creates hidden dependencies and does not allow to modify any of the components independently of the complete chain In general, all applications have common functional blocks like data publishers, transfer layer, data collectors, data repository, data processing engines, web service and UIs. SAM in addition has a test submission framework based on Nagios. 13-Sep-2013 27 NEC`2013, Varna, Julia Andreeva, CERN
Review of the current systems used for WLCG monitoring (technology)
NEC`2013, Varna, Julia Andreeva, CERN 13-Sep-2013 28
Conclusions of Elasticsearch evaluation for WLCG Transfer Dashboard
Elasticsearch 0.90.3 does not support grouping by terms of multiple fields for statistical aggregations.
Using Elasticsearch 0.90.3 for WLCG Transfer Monitoring we could achieve similar performance to 2 nd hit, i.e. cached, Oracle performance but this means using diverse workarounds for multi-field grouping.
Elasticsearch 1.0 includes a new Aggregation Module that will support grouping by terms of multiple fields for statistical aggregations Recommend re-evaluating ElasticSearch for WLCG Transfer Monitoring when 1.0 is available.
13-Sep-2013 29 NEC`2013, Varna, Julia Andreeva, CERN