Increased Complexity : Machines/vm sprawl Software failures gets harder to handle as machines number grow - Huge overheard, on development, - Planning for failure and maintenance gets harder - Not really developer friendly. Long cycles to deployments.
Google “Borg” .Unofficial name. Moved to a new system called Omega Bundles all the cluster’s machines into a single pooled view of resources. Evolved to Omega.
1. Run multiple types of distributed systems on the same cluster of machines, and dynamically share resources more eficiently. 2. Provide failure detection, task distribution, task tracking, task monitoring, low level resources management, and fine grained resources sharing. And scale to the thousands of nodes.
an application that runs distributed applications on Mesos. It has two components : The Scheduler : Interacts with Mesos, to book resources and launch tasks within the mesos slaves. Executors : gets information from the framework like environmental variables,configuration information and run tasks inside the mesos slaves.
an application that runs distributed applications on Mesos. It has two components : The Scheduler : Interacts with Mesos, to book resources and launch tasks within the mesos slaves. Executors : run tasks in mesos slaves. Mesos provides an API to build its own frameworks. And there are SDKs in mulitple languages.
resource isolation. Mesos containers are built around cgroups , to provide isolation of CPU, memory, disk … and namespaces for network isolation. Started leveraging containerirsation since 2010
Mesos Master uses Zookeeper for leader election and leader discovery. - Mesos Master have a Registrar to log all the tasks and slaves info, implemented as a replicated log using MultiPaxos for consensus
Mesos Master uses Zookeeper for leader election and leader discovery. - Mesos Master have a Registrar to log all the tasks and slaves info, implemented as a replicated log using MultiPaxos for consensus - Mesos has a slave recovery mechanism. So user’s tasks keep running whenever a slave proces goes under.
Mesos Master uses Zookeeper for leader election and leader discovery. - Mesos Master have a Registrar to log all the tasks and slaves info, implemented as a replicated log using MultiPaxos for consensus - Mesos has a slave recovery mechanism. So user’s tasks keep running whenever a slave proces goes under. Slaves checkpoint information to their local disks (task info, status updates, etc ), they recover checkpointed data and reconnect with old tasks when they restart.
an application that runs distributed applications on Mesos. It has two components : The Scheduler : Interacts with Mesos, to book resources and launch tasks within the mesos slaves. Executors : run tasks in mesos slaves. Mesos provides an API to build its own frameworks. And there are SDKs in mulitple languages.
to run long running services (web apps, etc …) - A distributed Init.d for the cluster. - Runs any linux binary without modification (Tomcat, Play, …) - Cluster wide process supervisor -Private PaaS - Service discovery - Provide a self service Rest API for deployment - Authentification & SSL - Placement constrainst (nodes, racks, etc. (Like with Spark) ) - Service discovery and load balancing via HAProxy What is Marathon ?