50,000 feet overview

• Transport: Jabber protocol
• Language: Python
• Probing: set of python modules + API
• Triggers: set of python modules + API
• Storage: local to nodes
• Query: API to query data on the node
• Monitor: mokup in GTK-Python

The big picture

Jabber for transport

• Instant messaging: permanent connection
• Offers group communication and point to point
• Homogeneous naming mechanism
• Open specification (IETF RFC)
• Free server with possibility of industrial solution
• Lazy: evaluation in 03/2001

Jabber Implementation

• Reimplemented minimal client capabilities in Python
• XML-RPC on top of Jabber for direct and group queries
• Encryption though SSL
• Authentication is weak ATM
• Code in CVS: rhn/playpen/Jabber

Extensible/Flexible infrastructure

• 5 Key APIs:
  • Probes: data acquisition
  • DB: data storage and query
  • Monitoring: checks for conditions and triggers events
  • Scheduler: on each client add/remove modules
  • Modules: dynamically query/check/load signed Python code
• Transparent remote invocation through XML-RPC/Jabber

Probes

• Base python class + derivation
• Sampling frequency can be modified
• Predefined: disk, load5, network use (and services ports)
• Probes are named e.g. "base/rhn/load5"
• Adding SNMP or custom samples should be easy (Python)

DB

• Only local storage ATM
• Using bsddb python interface to db (switch to DB3)
• API: list, last, closest, averages, avgList
• Auto cleanup by averaging after a month
• rpmfind since january: 40MB of data

Monitoring

• Module query the database for abnormal conditions
• Basically an entry point call periodically
• Anything scriptable in Python can be tested
• Dynamically removable
• Automatic dependancy to probes checked
• Geneate events which are logged and dispatched on the group
• Monitors are named too e.g. "rhn/base/netdev"

Scheduler

• Keep a list of modules to call and the interval
• Also check for incoming data from Jabber
• Handle the XML-RPC entry points

Modules

• PGP/GPG signed python modules
• uses popen3 to only load verified modules
• PBM: how to build the web ot trust (seed + delagation)
• Allow to extend both probes and monitors
• Module servers can be installed and queried for a given module

Monitor station

• Mokup in GTK-Python
• Shows basic interfaces, used to check the framework
• "nearly" usable
• I suck at UI work !!!

Monitor station screenshot

TODO List

• Improve the network code for disconnect
• Build an RHNAPP front-end connected to Jabber
• Database for sampling ? DB3 ? Alternate storage nodes
• Dynamic triggers: threshold on values and deltas
• Packaging is weak
• Testing, testing, testing