apache mesos
cluster management made easy
all things are distributed today:
- file systems
- databases
- caches
- services
- programming logic
and there are some reasons
- high availability
- fault tolerance
- performance
- no SPOFs (mostly)
- commodity hardware
but there are some proBlems
- nodes use to fail
- network splits
- non-trivial deployment
- extensive housekeeping
- resource management
- heterogenous nodes
- process parts use to fail
More about housekeeping
each time significant part of the system serves for monitoring and maintanance of critical components
- redundancy to avoid SPOFs
- but coordination is still needed
- leader election to avoid desync
- what if there is no quorum?
- static allocation of resources
- even virtual hosts fail
mesos concepts
- cluster or grid as a single resource
- distributed container for applications
- infinite scalability
- unified API for application developers
cluster operating system?
architecture
frameworks
cluster management made easy
By Anton Kirillov
