aurora-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
Subject svn commit: r1590066 [4/4] - in /incubator/aurora/site: ./ publish/documentation/latest/ publish/documentation/latest/client-commands/ publish/documentation/latest/clientcommands/ publish/documentation/latest/configuration-reference/ publish/documentat...
Date Fri, 25 Apr 2014 15:59:28 GMT
Added: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (added)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -0,0 +1,1139 @@
+Aurora Configuration Tutorial
+How to write Aurora configuration files, including feature descriptions
+and best practices. When writing a configuration file, make use of
+`aurora inspect`. It takes the same job key and configuration file
+arguments as `aurora create` or `aurora update`. It first ensures the
+configuration parses, then outputs it in human-readable form.
+You should read this after going through the general [Aurora Tutorial](/documentation/latest/tutorial/).
+[The Basics](#Basics)
+    [Use Bottom-To-Top Object Ordering](#Bottom)
+[An Example Configuration File](#Example)
+[Defining Process Objects](#Process)
+[Getting Your Code Into The Sandbox](#Sandbox)
+[Defining Task Objects](#Task)
+    [`SequentialTask`](#Sequential)
+    [`SimpleTask`](#Simple)
+    [`Tasks.concat` and `Tasks.combine`](#Concat)
+[Defining `Job` Objects](#Job)
+[Defining The `jobs` List](#jobs)
+[Templating 1: Binding in Pystachio](#Binding)
+[Structurals in Pystachio / Aurora](#Structurals)
+    [Mustaches Within Structurals](#Mustaches)
+[Templating 2: Structurals Are Factories](#Factories)
+    [A Second Way of Templating](#Second)
+[Advanced Binding](#AdvancedBinding)
+[Bind Syntax](#BindSyntax)
+    [Binding Complex Objects](#ComplexObjects)
+[Structural Binding](#StructuralBinding)
+[Configuration File Writing Tips And Best Practices](#Tips)
+    [Use As Few `.aurora` Files As Possible](#Few)
+    [Avoid Boilerplate](#Boilerplate)
+    [Thermos Uses bash, But Thermos Is Not bash](#Bash)
+    [Rarely Use Functions In Your Configurations](#Functions)
+The Basics
+To run a job on Aurora, you must specify a configuration file that tells
+Aurora what it needs to know to schedule the job, what Mesos needs to
+run the tasks the job is made up of, and what Thermos needs to run the
+processes that make up the tasks. This file must have
+a`.aurora` suffix.
+A configuration file defines a collection of objects, along with parameter
+values for their attributes. An Aurora configuration file contains the
+following three types of objects:
+- Job
+- Task
+- Process
+A configuration also specifies a list of `Job` objects assigned
+to the variable `jobs`.
+- jobs (list of defined Jobs to run)
+The `.aurora` file format is just Python. However, `Job`, `Task`,
+`Process`, and other classes are defined by a type-checked dictionary
+templating library called *Pystachio*, a powerful tool for
+configuration specification and reuse. Pystachio objects are tailored
+via {{}} surrounded templates.
+When writing your `.aurora` file, you may use any Pystachio datatypes, as
+well as any objects shown in the [*Aurora+Thermos Configuration
+Reference*](/documentation/latest/configuration-reference/), without `import` statements - the
+Aurora config loader injects them automatically. Other than that, an `.aurora`
+file works like any other Python script.
+[*Aurora+Thermos Configuration Reference*](/documentation/latest/configuration-reference/)
+has a full reference of all Aurora/Thermos defined Pystachio objects.
+### Use Bottom-To-Top Object Ordering
+A well-structured configuration starts with structural templates (if
+any). Structural templates encapsulate in their attributes all the
+differences between Jobs in the configuration that are not directly
+manipulated at the `Job` level, but typically at the `Process` or `Task`
+level. For example, if certain processes are invoked with slightly
+different settings or input.
+After structural templates, define, in order, `Process`es, `Task`s, and
+Structural template names should be *UpperCamelCased* and their
+instantiations are typically *UPPER\_SNAKE\_CASED*. `Process`, `Task`,
+and `Job` names are typically *lower\_snake\_cased*. Indentation is typically 2
+An Example Configuration File
+The following is a typical configuration file. Don't worry if there are
+parts you don't understand yet, but you may want to refer back to this
+as you read about its individual parts. Note that names surrounded by
+curly braces {{}} are template variables, which the system replaces with
+bound values for the variables.
+    # --- templates here ---
+	class Profile(Struct):
+	  package_version = Default(String, 'live')
+	  java_binary = Default(String, '/usr/lib/jvm/java-1.7.0-openjdk/bin/java')
+	  extra_jvm_options = Default(String, '')
+	  parent_environment = Default(String, 'prod')
+	  parent_serverset = Default(String,
+                                 '/foocorp/service/bird/{{parent_environment}}/bird')
+	# --- processes here ---
+	main = Process(
+	  name = 'application',
+	  cmdline = '{{profile.java_binary}} -server -Xmx1792m '
+	            '{{profile.extra_jvm_options}} '
+	            '-jar application.jar '
+	            '-upstreamService {{profile.parent_serverset}}'
+	)
+	# --- tasks ---
+	base_task = SequentialTask(
+	  name = 'application',
+	  processes = [
+	    Process(
+	      name = 'fetch',
+	      cmdline = 'curl -O
+        {{profile.package_version}}/application.jar'),
+	  ]
+	)
+        # not always necessary but often useful to have separate task
+        # resource classes
+        staging_task = base_task(resources =
+                         Resources(cpu = 1.0,
+                                   ram = 2048*MB,
+                                   disk = 1*GB))
+	production_task = base_task(resources =
+                            Resources(cpu = 4.0,
+                                      ram = 2560*MB,
+                                      disk = 10*GB))
+	# --- job template ---
+	job_template = Job(
+	  name = 'application',
+	  role = 'myteam',
+	  contact = '',
+	  instances = 20,
+	  service = True,
+	  task = production_task
+	)
+	# -- profile instantiations (if any) ---
+	PRODUCTION = Profile()
+	STAGING = Profile(
+	  extra_jvm_options = '-Xloggc:gc.log',
+	  parent_environment = 'staging'
+	)
+	# -- job instantiations --
+	jobs = [
+          job_template(cluster = 'cluster1', environment = 'prod')
+	               .bind(profile = PRODUCTION),
+          job_template(cluster = 'cluster2', environment = 'prod')
+	                .bind(profile = PRODUCTION),
+          job_template(cluster = 'cluster1',
+                        environment = 'staging',
+			service = False,
+			task = staging_task,
+			instances = 2)
+			.bind(profile = STAGING),
+	]
+## Defining Process Objects
+Processes are handled by the Thermos system. A process is a single
+executable step run as a part of an Aurora task, which consists of a
+bash-executable statement.
+The key (and required) `Process` attributes are:
+-   `name`: Any string which is a valid Unix filename (no slashes,
+    NULLs, or leading periods). The `name` value must be unique relative
+    to other Processes in a `Task`.
+-   `cmdline`: A command line run in a bash subshell, so you can use
+    bash scripts. Nothing is supplied for command-line arguments,
+    so `$*` is unspecified.
+Many tiny processes make managing configurations more difficult. For
+example, the following is a bad way to define processes.
+    copy = Process(
+      name = 'copy',
+      cmdline = 'curl -O'
+    )
+    unpack = Process(
+      name = 'unpack',
+      cmdline = 'unzip'
+    )
+    remove = Process(
+      name = 'remove',
+      cmdline = 'rm -f'
+    )
+    run = Process(
+      name = 'app',
+      cmdline = 'java -jar app.jar'
+    )
+    run_task = Task(
+      processes = [copy, unpack, remove, run],
+      constraints = order(copy, unpack, remove, run)
+    )
+Since `cmdline` runs in a bash subshell, you can chain commands
+with `&&` or `||`.
+When defining a `Task` that is just a list of Processes run in a
+particular order, use `SequentialTask`, as described in the [*Defining*
+`Task` *Objects*](#Task) section. The following simplifies and combines the
+above multiple `Process` definitions into just two.
+    stage = Process(
+      name = 'stage',
+      cmdline = 'curl -O && '
+                'unzip && rm -f')
+    run = Process(name = 'app', cmdline = 'java -jar app.jar')
+    run_task = SequentialTask(processes = [stage, run])
+`Process` also has five optional attributes, each with a default value
+if one isn't specified in the configuration:
+-   `max_failures`: Defaulting to `1`, the maximum number of failures
+    (non-zero exit statuses) before this `Process` is marked permanently
+    failed and not retried. If a `Process` permanently fails, Thermos
+    checks the `Process` object's containing `Task` for the task's
+    failure limit (usually 1) to determine whether or not the `Task`
+    should be failed. Setting `max_failures`to `0` means that this
+    process will keep retrying until a successful (zero) exit status is
+    achieved. Retries happen at most once every `min_duration` seconds
+    to prevent effectively mounting a denial of service attack against
+    the coordinating scheduler.
+-   `daemon`: Defaulting to `False`, if `daemon` is set to `True`, a
+    successful (zero) exit status does not prevent future process runs.
+    Instead, the `Process` reinvokes after `min_duration` seconds.
+    However, the maximum failure limit (`max_failures`) still
+    applies. A combination of `daemon=True` and `max_failures=0` retries
+    a `Process` indefinitely regardless of exit status. This should
+    generally be avoided for very short-lived processes because of the
+    accumulation of checkpointed state for each process run. When
+    running in Aurora, `max_failures` is capped at
+    100.
+-   `ephemeral`: Defaulting to `False`, if `ephemeral` is `True`, the
+    `Process`' status is not used to determine if its bound `Task` has
+    completed. For example, consider a `Task` with a
+    non-ephemeral webserver process and an ephemeral logsaver process
+    that periodically checkpoints its log files to a centralized data
+    store. The `Task` is considered finished once the webserver process
+    finishes, regardless of the logsaver's current status.
+-   `min_duration`: Defaults to `15`. Processes may succeed or fail
+    multiple times during a single Task. Each result is called a
+    *process run* and this value is the minimum number of seconds the
+    scheduler waits before re-running the same process.
+-   `final`: Defaulting to `False`, this is a finalizing `Process` that
+    should run last. Processes can be grouped into two classes:
+    *ordinary* and *finalizing*. By default, Thermos Processes are
+    ordinary. They run as long as the `Task` is considered
+    healthy (i.e. hasn't reached a failure limit). But once all regular
+    Thermos Processes have either finished or the `Task` has reached a
+    certain failure threshold, Thermos moves into a *finalization* stage
+    and runs all finalizing Processes. These are typically necessary for
+    cleaning up after the `Task`, such as log checkpointers, or perhaps
+    e-mail notifications of a completed Task. Finalizing processes may
+    not depend upon ordinary processes or vice-versa, however finalizing
+    processes may depend upon other finalizing processes and will
+    otherwise run as a typical process schedule.
+## Getting Your Code Into The Sandbox
+When using Aurora, you need to get your executable code into its "sandbox", specifically
+the Task sandbox where the code executes for the Processes that make up that Task.
+Each Task has a sandbox created when the Task starts and garbage
+collected when it finishes. All of a Task's processes run in its
+sandbox, so processes can share state by using a shared current
+working directory.
+Typically, you save this code somewhere. You then need to define a Process
+in your `.aurora` configuration file that fetches the code from that somewhere
+to where the slave can see it. For a public cloud, that can be anywhere public on
+the Internet, such as S3. For a private cloud internal storage, you need to put in
+on an accessible HDFS cluster or similar storage.
+The template for this Process is:
+    <name> = Process(
+      name = '<name>'
+      cmdline = '<command to copy and extract code archive into current working directory>'
+    )
+Note: Be sure the extracted code archive has an executable.
+## Defining Task Objects
+Tasks are handled by Mesos. A task is a collection of processes that
+runs in a shared sandbox. It's the fundamental unit Aurora uses to
+schedule the datacenter; essentially what Aurora does is find places
+in the cluster to run tasks.
+The key (and required) parts of a Task are:
+-   `name`: A string giving the Task's name. By default, if a Task is
+    not given a name, it inherits the first name in its Process list.
+-   `processes`: An unordered list of Process objects bound to the Task.
+    The value of the optional `constraints` attribute affects the
+    contents as a whole. Currently, the only constraint, `order`, determines if
+    the processes run in parallel or sequentially.
+-   `resources`: A `Resource` object defining the Task's resource
+        footprint. A `Resource` object has three attributes:
+        -   `cpu`: A Float, the fractional number of cores the Task
+        requires.
+        -   `ram`: An Integer, RAM bytes the Task requires.
+        -   `disk`: An integer, disk bytes the Task requires.
+A basic Task definition looks like:
+    Task(
+        name="hello_world",
+        processes=[Process(name = "hello_world", cmdline = "echo hello world")],
+        resources=Resources(cpu = 1.0,
+                            ram = 1*GB,
+                            disk = 1*GB))
+There are four optional Task attributes:
+-   `constraints`: A list of `Constraint` objects that constrain the
+    Task's processes. Currently there is only one type, the `order`
+    constraint. For example the following requires that the processes
+    run in the order `foo`, then `bar`.
+        constraints = [Constraint(order=['foo', 'bar'])]
+    There is an `order()` function that takes `order('foo', 'bar', 'baz')`
+    and converts it into `[Constraint(order=['foo', 'bar', 'baz'])]`.
+    `order()` accepts Process name strings `('foo', 'bar')` or the processes
+    themselves, e.g. `foo=Process(name='foo', ...)`, `bar=Process(name='bar', ...)`,
+    `constraints=order(foo, bar)`
+    Note that Thermos rejects tasks with process cycles.
+-   `max_failures`: Defaulting to `1`, the number of failed processes
+    needed for the `Task` to be marked as failed. Note how this
+    interacts with individual Processes' `max_failures` values. Assume a
+    Task has two Processes and a `max_failures` value of `2`. So both
+    Processes must fail for the Task to fail. Now, assume each of the
+    Task's Processes has its own `max_failures` value of `10`. If
+    Process "A" fails 5 times before succeeding, and Process "B" fails
+    10 times and is then marked as failing, their parent Task succeeds.
+    Even though there were 15 individual failures by its Processes, only
+    1 of its Processes was finally marked as failing. Since 1 is less
+    than the 2 that is the Task's `max_failures` value, the Task does
+    not fail.
+-   `max_concurrency`: Defaulting to `0`, the maximum number of
+    concurrent processes in the Task. `0` specifies unlimited
+    concurrency. For Tasks with many expensive but otherwise independent
+    processes, you can limit the amount of concurrency Thermos schedules
+    instead of artificially constraining them through `order`
+    constraints. For example, a test framework may generate a Task with
+    100 test run processes, but runs it in a Task with
+    `resources.cpus=4`. Limit the amount of parallelism to 4 by setting
+    `max_concurrency=4`.
+-   `finalization_wait`: Defaulting to `30`, the number of seconds
+    allocated for finalizing the Task's processes. A Task starts in
+    `ACTIVE` state when Processes run and stays there as long as the Task
+    is healthy and Processes run. When all Processes finish successfully
+    or the Task reaches its maximum process failure limit, it goes into
+    `CLEANING` state. In `CLEANING`, it sends `SIGTERMS` to any still running
+    Processes. When all Processes terminate, the Task goes into
+    `FINALIZING` state and invokes the schedule of all processes whose
+    final attribute has a True value. Everything from the end of `ACTIVE`
+    to the end of `FINALIZING` must happen within `finalization_wait`
+    number of seconds. If not, all still running Processes are sent
+    `SIGKILL`s (or if dependent on yet to be completed Processes, are
+    never invoked).
+### SequentialTask: Running Processes in Parallel or Sequentially
+By default, a Task with several Processes runs them in parallel. There
+are two ways to run Processes sequentially:
+-   Include an `order` constraint in the Task definition's `constraints`
+    attribute whose arguments specify the processes' run order:
+        Task( ... processes=[process1, process2, process3],
+	          constraints = order(process1, process2, process3), ...)
+-   Use `SequentialTask` instead of `Task`; it automatically runs
+    processes in the order specified in the `processes` attribute. No
+    `constraint` parameter is needed:
+        SequentialTask( ... processes=[process1, process2, process3] ...)
+### SimpleTask
+For quickly creating simple tasks, use the `SimpleTask` helper. It
+creates a basic task from a provided name and command line using a
+default set of resources. For example, in a .`aurora` configuration
+    SimpleTask(name="hello_world", command="echo hello world")
+is equivalent to
+    Task(name="hello_world",
+         processes=[Process(name = "hello_world", cmdline = "echo hello world")],
+         resources=Resources(cpu = 1.0,
+                             ram = 1*GB,
+                             disk = 1*GB))
+The simplest idiomatic Job configuration thus becomes:
+    import os
+    hello_world_job = Job(
+      task=SimpleTask(name="hello_world", command="echo hello world"),
+      role=os.getenv('USER'),
+      cluster="cluster1")
+When written to `hello_world.aurora`, you invoke it with a simple
+`aurora create cluster1/$USER/test/hello_world hello_world.aurora`.
+### Combining tasks
+`Tasks.concat`(synonym,`concat_tasks`) and
+`Tasks.combine`(synonym,`combine_tasks`) merge multiple Task definitions
+into a single Task. It may be easier to define complex Jobs
+as smaller constituent Tasks. But since a Job only includes a single
+Task, the subtasks must be combined before using them in a Job.
+Smaller Tasks can also be reused between Jobs, instead of having to
+repeat their definition for multiple Jobs.
+With both methods, the merged Task takes the first Task's name. The
+difference between the two is the result Task's process ordering.
+-   `Tasks.combine` runs its subtasks' processes in no particular order.
+    The new Task's resource consumption is the sum of all its subtasks'
+    consumption.
+-   `Tasks.concat` runs its subtasks in the order supplied, with each
+    subtask's processes run serially between tasks. It is analogous to
+    the `order` constraint helper, except at the Task level instead of
+    the Process level. The new Task's resource consumption is the
+    maximum value specified by any subtask for each Resource attribute
+    (cpu, ram and disk).
+For example, given the following:
+    setup_task = Task(
+      ...
+      processes=[download_interpreter, update_zookeeper],
+      # It is important to note that {{Tasks.concat}} has
+      # no effect on the ordering of the processes within a task;
+      # hence the necessity of the {{order}} statement below
+      # (otherwise, the order in which {{download_interpreter}}
+      # and {{update_zookeeper}} run will be non-deterministic)
+      constraints=order(download_interpreter, update_zookeeper),
+      ...
+    )
+    run_task = SequentialTask(
+      ...
+      processes=[download_application, start_application],
+      ...
+    )
+    combined_task = Tasks.concat(setup_task, run_task)
+The `Tasks.concat` command merges the two Tasks into a single Task and
+ensures all processes in `setup_task` run before the processes
+in `run_task`. Conceptually, the task is reduced to:
+    task = Task(
+      ...
+      processes=[download_interpreter, update_zookeeper,
+                 download_application, start_application],
+      constraints=order(download_interpreter, update_zookeeper,
+                        download_application, start_application),
+      ...
+    )
+In the case of `Tasks.combine`, the two schedules run in parallel:
+    task = Task(
+      ...
+      processes=[download_interpreter, update_zookeeper,
+                 download_application, start_application],
+      constraints=order(download_interpreter, update_zookeeper) +
+                        order(download_application, start_application),
+      ...
+    )
+In the latter case, each of the two sequences may operate in parallel.
+Of course, this may not be the intended behavior (for example, if
+the `start_application` Process implicitly relies
+upon `download_interpreter`). Make sure you understand the difference
+between using one or the other.
+## Defining Job Objects
+A job is a group of identical tasks that Aurora can run in a Mesos cluster.
+A `Job` object is defined by the values of several attributes, some
+required and some optional. The required attributes are:
+-   `task`: Task object to bind to this job. Note that a Job can
+    only take a single Task.
+-   `role`: Job's role account; in other words, the user account to run
+    the job as on a Mesos cluster machine. A common value is
+    `os.getenv('USER')`; using a Python command to get the user who
+    submits the job request. The other common value is the service
+    account that runs the job, e.g. `www-data`.
+-   `environment`: Job's environment, typical values
+    are `devel`, `test`, or `prod`.
+-   `cluster`: Aurora cluster to schedule the job in, defined in
+    `/etc/aurora/clusters.json` or `~/.clusters.json`. You can specify
+    jobs where the only difference is the `cluster`, then at run time
+    only run the Job whose job key includes your desired cluster's name.
+You usually see a `name` parameter. By default, `name` inherits its
+value from the Job's associated Task object, but you can override this
+default. For these four parameters, a Job definition might look like:
+    foo_job = Job( name = 'foo', cluster = 'cluster1',
+              role = os.getenv('USER'), environment = 'prod',
+              task = foo_task)
+In addition to the required attributes, there are several optional
+attributes. The first (strongly recommended) optional attribute is:
+-   `contact`: An email address for the Job's owner. For production
+    jobs, it is usually a team mailing list.
+Two more attributes deal with how to handle failure of the Job's Task:
+-   `max_task_failures`: An integer, defaulting to `1`, of the maximum
+    number of Task failures after which the Job is considered failed.
+    `-1` allows for infinite failures.
+-   `service`: A boolean, defaulting to `False`, which if `True`
+    restarts tasks regardless of whether they succeeded or failed. In
+    other words, if `True`, after the Job's Task completes, it
+    automatically starts again. This is for Jobs you want to run
+    continuously, rather than doing a single run.
+Three attributes deal with configuring the Job's Task:
+-   `instances`: Defaulting to `1`, the number of
+    instances/replicas/shards of the Job's Task to create.
+-   `priority`: Defaulting to `0`, the Job's Task's preemption priority,
+    for which higher values may preempt Tasks from Jobs with lower
+    values.
+-   `production`: a Boolean, defaulting to `False`, specifying that this
+    is a production job backed by quota. Tasks from production Jobs may
+    preempt tasks from any non-production job, and may only be preempted
+    by tasks from production jobs in the same role with higher
+    priority. **WARNING**: To run Jobs at this level, the Job role must
+    have the appropriate quota.
+The final three Job attributes each take an object as their value.
+-   `update_config`: An `UpdateConfig`
+    object provides parameters for controlling the rate and policy of
+    rolling updates. The `UpdateConfig` parameters are:
+    -   `batch_size`: An integer, defaulting to `1`, specifying the
+        maximum number of shards to update in one iteration.
+    -   `restart_threshold`: An integer, defaulting to `60`, specifying
+        the maximum number of seconds before a shard must move into the
+        `RUNNING` state before considered a failure.
+    -   `watch_secs`: An integer, defaulting to `30`, specifying the
+        minimum number of seconds a shard must remain in the `RUNNING`
+        state before considered a success.
+    -   `max_per_shard_failures`: An integer, defaulting to `0`,
+        specifying the maximum number of restarts per shard during an
+        update. When the limit is exceeded, it increments the total
+        failure count.
+    -   `max_total_failures`: An integer, defaulting to `0`, specifying
+        the maximum number of shard failures tolerated during an update.
+        Cannot be equal to or greater than the job's total number of
+        tasks.
+-   `health_check_config`: A `HealthCheckConfig` object that provides
+    parameters for controlling a Task's health checks via HTTP. Only
+    used if a health port was assigned with a command line wildcard. The
+    `HealthCheckConfig` parameters are:
+    -   `initial_interval_secs`: An integer, defaulting to `60`,
+        specifying the initial delay for doing an HTTP health check.
+    -   `interval_secs`: An integer, defaulting to `30`, specifying the
+        number of seconds in the interval between checking the Task's
+        health.
+    -   `timeout_secs`: An integer, defaulting to `1`, specifying the
+        number of seconds the application must respond to an HTTP health
+        check with `OK` before it is considered a failure.
+    -   `max_consecutive_failures`: An integer, defaulting to `0`,
+        specifying the maximum number of consecutive failures before a
+        task is unhealthy.
+-   `constraints`: A `dict` Python object, specifying Task scheduling
+    constraints. Most users will not need to specify constraints, as the
+    scheduler automatically inserts reasonable defaults. Please do not
+    set this field unless you are sure of what you are doing. See the
+    section in the Aurora + Thermos Reference manual on [Specifying
+    Scheduling Constraints](/documentation/latest/configuration-reference/) for more information.
+## The jobs List
+At the end of your `.aurora` file, you need to specify a list of the
+file's defined Jobs to run in the order listed. For example, the
+following runs first `job1`, then `job2`, then `job3`.
+jobs = [job1, job2, job3]
+The `.aurora` file format is just Python. However, `Job`, `Task`,
+`Process`, and other classes are defined by a templating library called
+*Pystachio*, a powerful tool for configuration specification and reuse.
+[Aurora+Thermos Configuration Reference](/documentation/latest/configuration-reference/)
+has a full reference of all Aurora/Thermos defined Pystachio objects.
+When writing your `.aurora` file, you may use any Pystachio datatypes, as
+well as any objects shown in the *Aurora+Thermos Configuration
+Reference* without `import` statements - the Aurora config loader
+injects them automatically. Other than that the `.aurora` format
+works like any other Python script.
+### Templating 1: Binding in Pystachio
+Pystachio uses the visually distinctive {{}} to indicate template
+variables. These are often called "mustache variables" after the
+similarly appearing variables in the Mustache templating system and
+because the curly braces resemble mustaches.
+If you are familiar with the Mustache system, templates in Pystachio
+have significant differences. They have no nesting, joining, or
+inheritance semantics. On the other hand, when evaluated, templates
+are evaluated iteratively, so this affords some level of indirection.
+Let's start with the simplest template; text with one
+variable, in this case `name`;
+    Hello {{name}}
+If we evaluate this as is, we'd get back:
+    Hello
+If a template variable doesn't have a value, when evaluated it's
+replaced with nothing. If we add a binding to give it a value:
+    { "name" : "Tom" }
+We'd get back:
+    Hello Tom
+We can also use {{}} variables as sectional variables. Let's say we
+    {{#x}} Testing... {{/x}}
+If `x` evaluates to `True`, the text between the sectional tags is
+shown. If there is no value for `x` or it evaluates to `False`, the
+between tags text is not shown. So, at a basic level, a sectional
+variable acts as a conditional.
+However, if the sectional variable evaluates to a list, array, etc. it
+acts as a `foreach`. For example,
+    {{#x}} {{name}} {{/x}}
+    { "x": [ { "name" : "tic" } { "name" : "tac" } { "name" : "toe" } ] }
+evaluates to
+    tic tac toe
+Every Pystachio object has an associated `.bind` method that can bind
+values to {{}} variables. Bindings are not immediately evaluated.
+Instead, they are evaluated only when the interpolated value of the
+object is necessary, e.g. for performing equality or serializing a
+message over the wire.
+Objects with and without mustache templated variables behave
+    >>> Float(1.5)
+    Float(1.5)
+    >>> Float('{{x}}.5')
+    Float({{x}}.5)
+    >>> Float('{{x}}.5').bind(x = 1)
+    Float(1.5)
+    >>> Float('{{x}}.5').bind(x = 1) == Float(1.5)
+    True
+    >>> contextual_object = String('{{metavar{{number}}}}').bind(
+    ... metavar1 = "first", metavar2 = "second")
+    >>> contextual_object
+    String({{metavar{{number}}}})
+    >>> contextual_object.bind(number = 1)
+    String(first)
+    >>> contextual_object.bind(number = 2)
+    String(second)
+You usually bind simple key to value pairs, but you can also bind three
+other objects: lists, dictionaries, and structurals. These will be
+described in detail later.
+### Structurals in Pystachio / Aurora
+Most Aurora/Thermos users don't ever (knowingly) interact with `String`,
+`Float`, or `Integer` Pystashio objects directly. Instead they interact
+with derived structural (`Struct`) objects that are collections of
+fundamental and structural objects. The structural object components are
+called *attributes*. Aurora's most used structural objects are `Job`,
+`Task`, and `Process`:
+    class Process(Struct):
+      cmdline = Required(String)
+      name = Required(String)
+      max_failures = Default(Integer, 1)
+      daemon = Default(Boolean, False)
+      ephemeral = Default(Boolean, False)
+      min_duration = Default(Integer, 5)
+      final = Default(Boolean, False)
+Construct default objects by following the object's type with (). If you
+want an attribute to have a value different from its default, include
+the attribute name and value inside the parentheses.
+    >>> Process()
+    Process(daemon=False, max_failures=1, ephemeral=False,
+      min_duration=5, final=False)
+Attribute values can be template variables, which then receive specific
+values when creating the object.
+    >>> Process(cmdline = 'echo {{message}}')
+    Process(daemon=False, max_failures=1, ephemeral=False, min_duration=5,
+            cmdline=echo {{message}}, final=False)
+    >>> Process(cmdline = 'echo {{message}}').bind(message = 'hello world')
+    Process(daemon=False, max_failures=1, ephemeral=False, min_duration=5,
+            cmdline=echo hello world, final=False)
+A powerful binding property is that all of an object's children inherit its
+    >>> List(Process)([
+    ... Process(name = '{{prefix}}_one'),
+    ... Process(name = '{{prefix}}_two')
+    ... ]).bind(prefix = 'hello')
+    ProcessList(
+      Process(daemon=False, name=hello_one, max_failures=1, ephemeral=False, min_duration=5, final=False),
+      Process(daemon=False, name=hello_two, max_failures=1, ephemeral=False, min_duration=5, final=False)
+      )
+Remember that an Aurora Job contains Tasks which contain Processes. A
+Job level binding is inherited by its Tasks and all their Processes.
+Similarly a Task level binding is available to that Task and its
+Processes but is *not* visible at the Job level (inheritance is a
+one-way street.)
+#### Mustaches Within Structurals
+When you define a `Struct` schema, one powerful, but confusing, feature
+is that all of that structure's attributes are Mustache variables within
+the enclosing scope *once they have been populated*.
+For example, when `Process` is defined above, all its attributes such as
+{{`name`}}, {{`cmdline`}}, {{`max_failures`}} etc., are all immediately
+defined as Mustache variables, implicitly bound into the `Process`, and
+inherit all child objects once they are defined.
+Thus, you can do the following:
+    >>> Process(name = "installer", cmdline = "echo {{name}} is running")
+    Process(daemon=False, name=installer, max_failures=1, ephemeral=False, min_duration=5,
+            cmdline=echo installer is running, final=False)
+WARNING: This binding only takes place in one direction. For example,
+the following does NOT work and does not set the `Process` `name`
+attribute's value.
+    >>> Process().bind(name = "installer")
+    Process(daemon=False, max_failures=1, ephemeral=False, min_duration=5, final=False)
+The following is also not possible and results in an infinite loop that
+attempts to resolve ``.
+    >>> Process(name = '{{name}}').bind(name = 'installer')
+Do not confuse Structural attributes with bound Mustache variables.
+Attributes are implicitly converted to Mustache variables but not vice
+### Templating 2: Structurals Are Factories
+#### A Second Way of Templating
+A second templating method is both as powerful as the aforementioned and
+often confused with it. This method is due to automatic conversion of
+Struct attributes to Mustache variables as described above.
+Suppose you create a Process object:
+    >>> p = Process(name = "process_one", cmdline = "echo hello world")
+    >>> p
+    Process(daemon=False, name=process_one, max_failures=1, ephemeral=False, min_duration=5,
+            cmdline=echo hello world, final=False)
+This `Process` object, "`p`", can be used wherever a `Process` object is
+needed. It can also be reused by changing the value(s) of its
+attribute(s). Here we change its `name` attribute from `process_one` to
+    >>> p(name = "process_two")
+    Process(daemon=False, name=process_two, max_failures=1, ephemeral=False, min_duration=5,
+            cmdline=echo hello world, final=False)
+Template creation is a common use for this technique:
+    >>> Daemon = Process(daemon = True)
+    >>> logrotate = Daemon(name = 'logrotate', cmdline = './logrotate conf/logrotate.conf')
+    >>> mysql = Daemon(name = 'mysql', cmdline = 'bin/mysqld --safe-mode')
+### Advanced Binding
+As described above, `.bind()` binds simple strings or numbers to
+Mustache variables. In addition to Structural types formed by combining
+atomic types, Pystachio has two container types; `List` and `Map` which
+can also be bound via `.bind()`.
+#### Bind Syntax
+The `bind()` function can take Python dictionaries or `kwargs`
+interchangeably (when "`kwargs`" is in a function definition, `kwargs`
+receives a Python dictionary containing all keyword arguments after the
+formal parameter list).
+    >>> String('{{foo}}').bind(foo = 'bar') == String('{{foo}}').bind({'foo': 'bar'})
+    True
+Bindings done "closer" to the object in question take precedence:
+    >>> p = Process(name = '{{context}}_process')
+    >>> t = Task().bind(context = 'global')
+    >>> t(processes = [p, p.bind(context = 'local')])
+    Task(processes=ProcessList(
+      Process(daemon=False, name=global_process, max_failures=1, ephemeral=False, final=False,
+              min_duration=5),
+      Process(daemon=False, name=local_process, max_failures=1, ephemeral=False, final=False,
+              min_duration=5)
+    ))
+#### Binding Complex Objects
+##### Lists
+    >>> fibonacci = List(Integer)([1, 1, 2, 3, 5, 8, 13])
+    >>> String('{{fib[4]}}').bind(fib = fibonacci)
+    String(5)
+##### Maps
+    >>> first_names = Map(String, String)({'Kent': 'Clark', 'Wayne': 'Bruce', 'Prince': 'Diana'})
+    >>> String('{{first[Kent]}}').bind(first = first_names)
+    String(Clark)
+##### Structurals
+    >>> String('{{p.cmdline}}').bind(p = Process(cmdline = "echo hello world"))
+    String(echo hello world)
+### Structural Binding
+Use structural templates when binding more than two or three individual
+values at the Job or Task level. For fewer than two or three, standard
+key to string binding is sufficient.
+Structural binding is a very powerful pattern and is most useful in
+Aurora/Thermos for doing Structural configuration. For example, you can
+define a job profile. The following profile uses `HDFS`, the Hadoop
+Distributed File System, to designate a file's location. `HDFS` does
+not come with Aurora, so you'll need to either install it separately
+or change the way the dataset is designated.
+    class Profile(Struct):
+      version = Required(String)
+      environment = Required(String)
+      dataset = Default(String, hdfs://home/aurora/data/{{environment}}')
+    PRODUCTION = Profile(version = 'live', environment = 'prod')
+    DEVEL = Profile(version = 'latest',
+                    environment = 'devel',
+                    dataset = 'hdfs://home/aurora/data/test')
+    TEST = Profile(version = 'latest', environment = 'test')
+    JOB_TEMPLATE = Job(
+      name = 'application',
+      role = 'myteam',
+      cluster = 'cluster1',
+      environment = '{{profile.environment}}',
+      task = SequentialTask(
+        name = 'task',
+        resources = Resources(cpu = 2, ram = 4*GB, disk = 8*GB),
+        processes = [
+	  Process(name = 'main', cmdline = 'java -jar application.jar -hdfsPath
+                 {{profile.dataset}}')
+        ]
+       )
+     )
+    jobs = [
+      JOB_TEMPLATE(instances = 100).bind(profile = PRODUCTION),
+      JOB_TEMPLATE.bind(profile = DEVEL),
+      JOB_TEMPLATE.bind(profile = TEST),
+     ]
+In this case, a custom structural "Profile" is created to self-document
+the configuration to some degree. This also allows some schema
+"type-checking", and for default self-substitution, e.g. in
+`Profile.dataset` above.
+So rather than a `.bind()` with a half-dozen substituted variables, you
+can bind a single object that has sensible defaults stored in a single
+Configuration File Writing Tips And Best Practices
+### Use As Few .aurora Files As Possible
+When creating your `.aurora` configuration, try to keep all versions of
+a particular job within the same `.aurora` file. For example, if you
+have separate jobs for `cluster1`, `cluster1` staging, `cluster1`
+testing, and`cluster2`, keep them as close together as possible.
+Constructs shared across multiple jobs owned by your team (e.g.
+team-level defaults or structural templates) can be split into separate
+`.aurora`files and included via the `include` directive.
+### Avoid Boilerplate
+If you see repetition or find yourself copy and pasting any parts of
+your configuration, it's likely an opportunity for templating. Take the
+example below:
+`redundant.aurora` contains:
+    download = Process(
+      name = 'download',
+      cmdline = 'wget',
+      max_failures = 5,
+      min_duration = 1)
+    unpack = Process(
+      name = 'unpack',
+      cmdline = 'rm -rf Python-2.7.3 && tar xzf Python-2.7.3.tar.bz2',
+      max_failures = 5,
+      min_duration = 1)
+    build = Process(
+      name = 'build',
+      cmdline = 'pushd Python-2.7.3 && ./configure && make && popd',
+      max_failures = 1)
+    email = Process(
+      name = 'email',
+      cmdline = 'echo Success | mail',
+      max_failures = 5,
+      min_duration = 1)
+    build_python = Task(
+      name = 'build_python',
+      processes = [download, unpack, build, email],
+      constraints = [Constraint(order = ['download', 'unpack', 'build', 'email'])])
+As you'll notice, there's a lot of repetition in the `Process`
+definitions. For example, almost every process sets a `max_failures`
+limit to 5 and a `min_duration` to 1. This is an opportunity for factoring
+into a common process template.
+Furthermore, the Python version is repeated everywhere. This can be
+bound via structural templating as described in the [Advanced Binding](#AdvancedBinding)
+`less_redundant.aurora` contains:
+    class Python(Struct):
+      version = Required(String)
+      base = Default(String, 'Python-{{version}}')
+      package = Default(String, '{{base}}.tar.bz2')
+    ReliableProcess = Process(
+      max_failures = 5,
+      min_duration = 1)
+    download = ReliableProcess(
+      name = 'download',
+      cmdline = 'wget{{python.version}}/{{python.package}}')
+    unpack = ReliableProcess(
+      name = 'unpack',
+      cmdline = 'rm -rf {{python.base}} && tar xzf {{python.package}}')
+    build = ReliableProcess(
+      name = 'build',
+      cmdline = 'pushd {{python.base}} && ./configure && make && popd',
+      max_failures = 1)
+    email = ReliableProcess(
+      name = 'email',
+      cmdline = 'echo Success | mail {{role}}')
+    build_python = SequentialTask(
+      name = 'build_python',
+      processes = [download, unpack, build, email]).bind(python = Python(version = "2.7.3"))
+### Thermos Uses bash, But Thermos Is Not bash
+#### Bad
+Many tiny Processes makes for harder to manage configurations.
+    copy = Process(
+      name = 'copy',
+      cmdline = 'rcp user@my_machine:my_application .'
+     )
+     unpack = Process(
+       name = 'unpack',
+       cmdline = 'unzip'
+     )
+     remove = Process(
+       name = 'remove',
+       cmdline = 'rm -f'
+     )
+     run = Process(
+       name = 'app',
+       cmdline = 'java -jar app.jar'
+     )
+     run_task = Task(
+       processes = [copy, unpack, remove, run],
+       constraints = order(copy, unpack, remove, run)
+     )
+#### Good
+Each `cmdline` runs in a bash subshell, so you have the full power of
+bash. Chaining commands with `&&` or `||` is almost always the right
+thing to do.
+Also for Tasks that are simply a list of processes that run one after
+another, consider using the `SequentialTask` helper which applies a
+linear ordering constraint for you.
+    stage = Process(
+      name = 'stage',
+      cmdline = 'rcp user@my_machine:my_application . && unzip && rm -f')
+    run = Process(name = 'app', cmdline = 'java -jar app.jar')
+    run_task = SequentialTask(processes = [stage, run])
+### Rarely Use Functions In Your Configurations
+90% of the time you define a function in a `.aurora` file, you're
+probably Doing It Wrong(TM).
+#### Bad
+    def get_my_task(name, user, cpu, ram, disk):
+      return Task(
+        name = name,
+        user = user,
+        processes = [STAGE_PROCESS, RUN_PROCESS],
+        constraints = order(STAGE_PROCESS, RUN_PROCESS),
+        resources = Resources(cpu = cpu, ram = ram, disk = disk)
+     )
+     task_one = get_my_task('task_one', 'feynman', 1.0, 32*MB, 1*GB)
+     task_two = get_my_task('task_two', 'feynman', 2.0, 64*MB, 1*GB)
+#### Good
+This one is more idiomatic. Forced keyword arguments prevents accidents,
+e.g. constructing a task with "32*MB" when you mean 32MB of ram and not
+disk. Less proliferation of task-construction techniques means
+easier-to-read, quicker-to-understand, and a more composable
+    TASK_TEMPLATE = SequentialTask(
+      user = 'wickman',
+      processes = [STAGE_PROCESS, RUN_PROCESS],
+    )
+    task_one = TASK_TEMPLATE(
+      name = 'task_one',
+      resources = Resources(cpu = 1.0, ram = 32*MB, disk = 1*GB) )
+    task_two = TASK_TEMPLATE(
+      name = 'task_two',
+      resources = Resources(cpu = 2.0, ram = 64*MB, disk = 1*GB)
+    )

Modified: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (original)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -81,25 +81,12 @@ In a cluster with `N` schedulers, the fl
 `floor(N/2) + 1`. So in a cluster with 1 scheduler it should be set to `1`, in a cluster with 3 it
 should be set to `2`, and in a cluster of 5 it should be set to `3`.
-  <thead>
-    <tr>
-      <th>Number of schedulers (N)
-      <th><code>-native_log_quorum_size</code> setting (<code>floor(N/2) + 1</code>)
-  <tbody>
-    <tr>
-      <td>1
-      <td>1
-    <tr>
-      <td>3
-      <td>2
-    <tr>
-      <td>5
-      <td>3
-    <tr>
-      <td>7
-      <td>4
+  Number of schedulers (N) | ```-native_log_quorum_size``` setting (```floor(N/2) + 1```)
+  ------------------------ | -------------------------------------------------------------
+  1                        | 1
+  3                        | 2
+  5                        | 3
+  7                        | 4
 *Incorrectly setting this flag will cause data corruption to occur!*

Modified: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (original)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -84,39 +84,13 @@ Note that the `post_` and `err_` hooks f
 You can associate `pre_`, `post_`, and `err_` hooks with the following methods. Since you do not directly interact with the methods, but rather the Aurora Command Line commands that call them, for each method we also list the command(s) that can call the method. Note that a different method or methods may be called by a command depending on how the command's other code executes. Similarly, multiple commands can call the same method. We also list the methods' argument signatures, which are used by their associated hooks. <a name="Chart"></a>
-<table border="1" cellpadding="0" cellspacing="0">
-  <tbody>
-    <tr>
-      <th>Aurora Client API Method</td>
-      <th>Client API Method Argument Signature</td>
-      <th>Aurora Command Line Command</td>
-    </tr>
-    <tr>
-      <td><code>cancel_update</code></td>
-      <td><code>self</code>, <code>job_key</code></td>
-      <td><code>cancel_update</code></td>
-    </tr>
-    <tr>
-      <td><code>create_job</code></td>
-      <td><code>self</code>, <code>config</code></td>
-      <td><code>create</code>, <code>runtask</code></td>
-    </tr>
-    <tr>
-      <td><code>restart</code></td>
-      <td><code>self</code>, <code>job_key</code>, <code>shards</code>, <code>update_config</code>, <code>health_check_interval_seconds</code></td>
-      <td><code>restart</code></td>
-    </tr>
-    <tr>
-      <td><code>update_job</code></td>
-      <td><code>self</code>, <code>config</code>, <code>health_check_interval_seconds=3</code>, <code>shards=None</code></td>
-      <td><code>update</code></td>
-     </tr>
-     <tr>
-       <td><code>kill_job</code></td>
-       <td><code>self</code>, <code>job_key</code>, <code>shards=None</code></td>
-       <td><code>kill</code></td>
-     </tr>
-   </table>
+  Aurora Client API Method | Client API Method Argument Signature | Aurora Command Line Command
+  -------------------------| ------------------------------------- | ---------------------------
+  ```cancel_update``` | ```self```, ```job_key``` | ```cancel_update```
+  ```create_job``` | ```self```, ```config``` | ```create```, <code>runtask
+  ```restart``` | ```self```, ```job_key```, ```shards```, ```update_config```, ```health_check_interval_seconds``` | ```restart```
+  ```update_job``` | ```self```, ```config```, ```health_check_interval_seconds=3```, ```shards=None``` | ```update```
+  ```kill_job``` | ```self```, ```job_key```, ```shards=None``` |  ```kill```
 Some specific examples:

Added: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (added)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -0,0 +1,147 @@
+Resource Isolation and Sizing
+**NOTE**: Resource Isolation and Sizing is very much a work in progress.
+Both user-facing aspects and how it works under the hood are subject to
+- [Introduction](#introduction)
+- [CPU Isolation](#cpu-isolation)
+- [CPU Sizing](#cpu-sizing)
+- [Memory Isolation](#memory-isolation)
+- [Memory Sizing](#memory-sizing)
+- [Disk Space](#disk-space)
+- [Disk Space Sizing](#disk-space-sizing)
+- [Other Resources](#other-resources)
+## Introduction
+Aurora is a multi-tenant system; a single software instance runs on a
+server, serving multiple clients/tenants. To share resources among
+tenants, it implements isolation of:
+* CPU
+* memory
+* disk space
+CPU is a soft limit, and handled differently from memory and disk space.
+Too low a CPU value results in throttling your application and
+slowing it down. Memory and disk space are both hard limits; when your
+application goes over these values, it's killed.
+Let's look at each resource type in more detail:
+## CPU Isolation
+Mesos uses a quota based CPU scheduler (the *Completely Fair Scheduler*)
+to provide consistent and predictable performance.  This is effectively
+a guarantee of resources -- you receive at least what you requested, but
+also no more than you've requested.
+The scheduler gives applications a CPU quota for every 100 ms interval.
+When an application uses its quota for an interval, it is throttled for
+the rest of the 100 ms. Usage resets for each interval and unused
+quota does not carry over.
+For example, an application specifying 4.0 CPU has access to 400 ms of
+CPU time every 100 ms. This CPU quota can be used in different ways,
+depending on the application and available resources. Consider the
+scenarios shown in this diagram.
+![CPU Availability](images/CPUavailability.png)
+* *Scenario A*: the application can use up to 4 cores continuously for
+every 100 ms interval. It is never throttled and starts processing
+new requests immediately.
+* *Scenario B* : the application uses up to 8 cores (depending on
+availability) but is throttled after 50 ms. The CPU quota resets at the
+start of each new 100 ms interval.
+* *Scenario C* : is like Scenario A, but there is a garbage collection
+event in the second interval that consumes all CPU quota. The
+application throttles for the remaining 75 ms of that interval and
+cannot service requests until the next interval. In this example, the
+garbage collection finished in one interval but, depending on how much
+garbage needs collecting, it may take more than one interval and further
+delay service of requests.
+*Technical Note*: Mesos considers logical cores, also known as
+hyperthreading or SMT cores, as the unit of CPU.
+## CPU Sizing
+To correctly size Aurora-run Mesos tasks, specify a per-shard CPU value
+that lets the task run at its desired performance when at peak load
+distributed across all shards. Include reserve capacity of at least 50%,
+possibly more, depending on how critical your service is (or how
+confident you are about your original estimate : -)), ideally by
+increasing the number of shards to also improve resiliency. When running
+your application, observe its CPU stats over time. If consistently at or
+near your quota during peak load, you should consider increasing either
+per-shard CPU or the number of shards.
+## Memory Isolation
+Mesos uses dedicated memory allocation. Your application always has
+access to the amount of memory specified in your configuration. The
+application's memory use is defined as the sum of the resident set size
+(RSS) of all processes in a shard. Each shard is considered
+In other words, say you specified a memory size of 10GB. Each shard
+would receive 10GB of memory. If an individual shard's memory demands
+exceed 10GB, that shard is killed, but the other shards continue
+*Technical note*: Total memory size is not enforced at allocation time,
+so your application can request more than its allocation without getting
+an ENOMEM. However, it will be killed shortly after.
+## Memory Sizing
+Size for your application's peak requirement. Observe the per-instance
+memory statistics over time, as memory requirements can vary over
+different periods. Remember that if your application exceeds its memory
+value, it will be killed, so you should also add a safety margin of
+around 10-20%. If you have the ability to do so, you may also want to
+put alerts on the per-instance memory.
+## Disk Space
+Disk space used by your application is defined as the sum of the files'
+disk space in your application's directory, including the `stdout` and
+`stderr` logged from your application. Each shard is considered
+independently. You should use off-node storage for your application's
+data whenever possible.
+In other words, say you specified disk space size of 100MB. Each shard
+would receive 100MB of disk space. If an individual shard's disk space
+demands exceed 100MB, that shard is killed, but the other shards
+continue working.
+After your application finishes running, its allocated disk space is
+reclaimed. Thus, your job's final action should move any disk content
+that you want to keep, such as logs, to your home file system or other
+less transitory storage. Disk reclamation takes place an undefined
+period after the application finish time; until then, the disk contents
+are still available but you shouldn't count on them being so.
+*Technical note* : Disk space is not enforced at write so your
+application can write above its quota without getting an ENOSPC, but it
+will be killed shortly after. This is subject to change.
+## Disk Space Sizing
+Size for your application's peak requirement. Rotate and discard log
+files as needed to stay within your quota. When running a Java process,
+add the maximum size of the Java heap to your disk space requirement, in
+order to account for an out of memory error dumping the heap
+into the application's sandbox space.
+## Other Resources
+Other resources, such as network bandwidth, do not have any performance
+guarantees. For some resources, such as memory bandwidth, there are no
+practical sharing methods so some application combinations collocated on
+the same host may cause contention.

Modified: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (original)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -1,9 +1,6 @@
 Aurora Tutorial
-Before reading this document, you should read over the (short) [README](/documentation/latest/README/)
-for the Aurora docs.
 - [Introduction](#introduction)
 - [Setup: Install Aurora](#setup-install-aurora)
 - [The Script](#the-script)
@@ -34,7 +31,8 @@ To get help, email questions to the Auro
 You use the Aurora client and web UI to interact with Aurora jobs. To
 install it locally, see [](/documentation/latest/vagrant/). The remainder of this
-Tutorial assumes you are running Aurora using Vagrant.
+Tutorial assumes you are running Aurora using Vagrant.  Unless otherwise stated,
+all commands are to be run from the root of the aurora repository clone.
 ## The Script
@@ -45,6 +43,8 @@ this directory is the same as `/vagrant`
 The script has an intentional bug, which we will explain later on.
+<!-- NOTE: If you are changing this file, be sure to also update examples/vagrant/
 import sys
 import time
@@ -66,17 +66,24 @@ if __name__ == "__main__":
 Once we have our script/program, we need to create a *configuration
 file* that tells Aurora how to manage and launch our Job. Save the below
-code in the file `hello_world.aurora` in the same directory as your
-`` file. (all Aurora configuration files end with `.aurora` and
-are written in a Python variant).
+code in the file `hello_world.aurora`.
+<!-- NOTE: If you are changing this file, be sure to also update examples/vagrant/
-import os
+pkg_path = '/vagrant/'
+# we use a trick here to make the configuration change with
+# the contents of the file, for simplicity.  in a normal setting, packages would be
+# versioned, and the version number would be changed in the configuration.
+import hashlib
+with open(pkg_path, 'rb') as f:
+  pkg_checksum = hashlib.md5(
 # copy into the local sandbox
 install = Process(
   name = 'fetch_package',
-  cmdline = 'cp /vagrant/ . && chmod +x')
+  cmdline = 'cp %s . && echo %s && chmod +x' % (pkg_path, pkg_checksum))
 # run the script
 hello_world = Process(
@@ -89,14 +96,17 @@ hello_world_task = SequentialTask(
   resources = Resources(cpu = 1, ram = 1*MB, disk=8*MB))
 jobs = [
-  Job(name = 'hello_world', cluster = 'example', role = 'www-data',
-      environment = 'devel', task = hello_world_task)
+  Service(cluster = 'devcluster',
+          environment = 'devel',
+          role = 'www-data',
+          name = 'hello_world',
+          task = hello_world_task)
 For more about Aurora configuration files, see the [Configuration
-Tutorial](/documentation/latest/configurationtutorial/) and the [Aurora + Thermos
-Reference](/documentation/latest/configurationreference/) (preferably after finishing this
+Tutorial](/documentation/latest/configuration-tutorial/) and the [Aurora + Thermos
+Reference](/documentation/latest/configuration-reference/) (preferably after finishing this
 ## What's Going On In That Configuration File?
@@ -132,7 +142,7 @@ identical, the job keys identify the sam
 cluster names. For Vagrant, from the top-level of your Aurora repository clone,
-    $ vagrant ssh aurora-scheduler
+    $ vagrant ssh
 Followed by:
@@ -142,8 +152,8 @@ You'll see something like:
-  "name": "example",
-  "zk": "",
+  "name": "devcluster",
+  "zk": "",
   "scheduler_zk_path": "/aurora/scheduler",
   "auth_mechanism": "UNAUTHENTICATED"
@@ -163,20 +173,17 @@ specified by its job key and configurati
 Or for our example:
-    aurora create example/www-data/devel/hello_world /vagrant/hello_world.aurora
-Note: Remember, the job key's `<jobname>` value is the name of the Job, not the name
-of its code file.
+    aurora create devcluster/www-data/devel/hello_world /vagrant/hello_world.aurora
 This returns:
-    $ vagrant ssh aurora-scheduler
+    $ vagrant ssh
     Welcome to Ubuntu 12.04 LTS (GNU/Linux 3.2.0-23-generic x86_64)
      * Documentation:
     Welcome to your Vagrant-built virtual machine.
     Last login: Fri Jan  3 02:18:55 2014 from
-    vagrant@precise64:~$ aurora create example/www-data/devel/hello_world \
+    vagrant@precise64:~$ aurora create devcluster/www-data/devel/hello_world \
      INFO] Creating job hello_world
      INFO] Response from scheduler: OK (message: 1 new tasks pending for job
@@ -187,7 +194,7 @@ This returns:
 Now that our job is running, let's see what it's doing. Access the
 scheduler web interface at `http://$scheduler_hostname:$scheduler_port/scheduler`
-Or when using `vagrant`, ``
+Or when using `vagrant`, ``
 First we see what Jobs are scheduled:
 ![Scheduled Jobs](images/ScheduledJobs.png)
@@ -197,15 +204,14 @@ with that role:
 ![Role Jobs](images/RoleJobs.png)
-Uh oh, that `Unstable` next to our `hello_world` Job doesn't look good. Click the
-`hello_world` Job, and you'll see:
+If you click on your `hello_world` Job, you'll see:
 ![hello_world Job](images/HelloWorldJob.png)
 Oops, looks like our first job didn't quite work! The task failed, so we have
 to figure out what went wrong.
-Access the page for our Task by clicking on its Host.
+Access the page for our Task by clicking on its host.
 ![Task page](images/TaskBreakdown.png)
@@ -217,12 +223,10 @@ to `stderr` on the failed `hello_world` 
 ![stderr page](images/stderr.png)
 It looks like we made a typo in our Python script. We wanted `xrange`,
-not `xrang`. Edit the `` script, save as `` and change your
-`hello_world.aurora` config file to use `` instead of ``.
-Now that we've updated our configuration, let's restart the job:
+not `xrang`. Edit the `` script to use the correct function and
+we will try again.
-    aurora update example/www-data/devel/hello_world /vagrant/hello_world.aurora
+    aurora update devcluster/www-data/devel/hello_world /vagrant/hello_world.aurora
 This time, the task comes up, we inspect the page, and see that the
 `hello_world` process is running.
@@ -238,13 +242,13 @@ output:
 Now that we're done, we kill the job using the Aurora client:
-    vagrant@precise64:~$ aurora kill example/www-data/devel/hello_world
-     INFO] Killing tasks for job: example/www-data/devel/hello_world
+    vagrant@precise64:~$ aurora killall devcluster/www-data/devel/hello_world
+     INFO] Killing tasks for job: devcluster/www-data/devel/hello_world
      INFO] Response from scheduler: OK (message: Tasks killed.)
      INFO] Job url: http://precise64:8081/scheduler/www-data/devel/hello_world
-The Task scheduler page now shows the `hello_world` process as `KILLED`.
+The job page now shows the `hello_world` tasks as completed.
 ![Killed Task page](images/killedtask.png)
@@ -252,10 +256,10 @@ The Task scheduler page now shows the `h
 Now that you've finished this Tutorial, you should read or do the following:
-- [The Aurora Configuration Tutorial](/documentation/latest/configurationtutorial/), which provides more examples
+- [The Aurora Configuration Tutorial](/documentation/latest/configuration-tutorial/), which provides more examples
   and best practices for writing Aurora configurations. You should also look at
-  the [Aurora + Thermos Configuration Reference](/documentation/latest/configurationreference/).
-- The [Aurora User Guide](/documentation/latest/userguide/) provides an overview of how Aurora, Mesos, and
+  the [Aurora + Thermos Configuration Reference](/documentation/latest/configuration-reference/).
+- The [Aurora User Guide](/documentation/latest/user-guide/) provides an overview of how Aurora, Mesos, and
   Thermos work "under the hood".
 - Explore the Aurora Client - use the `aurora help` subcommand, and read the
-  [Aurora Client Commands](/documentation/latest/clientcommands/) document.
+  [Aurora Client Commands](/documentation/latest/client-commands/) document.

Added: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (added)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -0,0 +1,292 @@
+Aurora User Guide
+- [Overview](#overview)
+- [Job Lifecycle](#job-lifecycle)
+  - [Life Of A Task](#life-of-a-task)
+  - [PENDING to RUNNING states](#pending-to-running-states)
+  - [Task Updates](#task-updates)
+  - [Giving Priority to Production Tasks: PREEMPTING](#giving-priority-to-production-tasks-preempting)
+  - [Natural Termination: FINISHED, FAILED](#natural-termination-finished-failed)
+  - [Forceful Termination: KILLING, RESTARTING](#forceful-termination-killing-restarting)
+- [Configuration](#configuration)
+- [Creating Jobs](#creating-jobs)
+- [Interacting With Jobs](#interacting-with-jobs)
+This document gives an overview of how Aurora works under the hood.
+It assumes you've already worked through the "hello world" example
+job in the [Aurora Tutorial](/documentation/latest/tutorial/). Specifics of how to use Aurora are **not**
+ given here, but pointers to documentation about how to use Aurora are
+Aurora is a Mesos framework used to schedule *jobs* onto Mesos. Mesos
+cares about individual *tasks*, but typical jobs consist of dozens or
+hundreds of task replicas. Aurora provides a layer on top of Mesos with
+its `Job` abstraction. An Aurora `Job` consists of a task template and
+instructions for creating near-identical replicas of that task (modulo
+things like "instance id" or specific port numbers which may differ from
+machine to machine).
+How many tasks make up a Job is complicated. On a basic level, a Job consists of
+one task template and instructions for creating near-idential replicas of that task
+(otherwise referred to as "instances" or "shards").
+However, since Jobs can be updated on the fly, a single Job identifier or *job key*
+can have multiple job configurations associated with it.
+For example, consider when I have a Job with 4 instances that each
+request 1 core of cpu, 1 GB of RAM, and 1 GB of disk space as specified
+in the configuration file `hello_world.aurora`. I want to
+update it so it requests 2 GB of RAM instead of 1. I create a new
+configuration file to do that called `new_hello_world.aurora` and
+issue a `aurora update --shards=0-1 <job_key_value> new_hello_world.aurora`
+This results in instances 0 and 1 having 1 cpu, 2 GB of RAM, and 1 GB of disk space,
+while instances 2 and 3 have 1 cpu, 1 GB of RAM, and 1 GB of disk space. If instance 3
+dies and restarts, it restarts with 1 cpu, 1 GB RAM, and 1 GB disk space.
+So that means there are two simultaneous task configurations for the same Job
+at the same time, just valid for different ranges of instances.
+This isn't a recommended pattern, but it is valid and supported by the
+Aurora scheduler. This most often manifests in the "canary pattern" where
+instance 0 runs with a different configuration than instances 1-N to test
+different code versions alongside the actual production job.
+A task can merely be a single *process* corresponding to a single
+command line, such as `python2.6`. However, a task can also
+consist of many separate processes, which all run within a single
+sandbox. For example, running multiple cooperating agents together,
+such as `logrotate`, `installer`, master, or slave processes. This is
+where Thermos  comes in. While Aurora provides a `Job` abstraction on
+top of Mesos `Tasks`, Thermos provides a `Process` abstraction
+underneath Mesos `Task`s and serves as part of the Aurora framework's
+You define `Job`s,` Task`s, and `Process`es in a configuration file.
+Configuration files are written in Python, and make use of the Pystachio
+templating language. They end in a `.aurora` extension.
+Pystachio is a type-checked dictionary templating library.
+> TL;DR
+> -   Aurora manages jobs made of tasks.
+> -   Mesos manages tasks made of processes.
+> -   Thermos manages processes.
+> -   All defined in `.aurora` configuration file.
+![Aurora hierarchy](images/aurora_hierarchy.png)
+Each `Task` has a *sandbox* created when the `Task` starts and garbage
+collected when it finishes. All of a `Task'`s processes run in its
+sandbox, so processes can share state by using a shared current working
+The sandbox garbage collection policy considers many factors, most
+importantly age and size. It makes a best-effort attempt to keep
+sandboxes around as long as possible post-task in order for service
+owners to inspect data and logs, should the `Task` have completed
+abnormally. But you can't design your applications assuming sandboxes
+will be around forever, e.g. by building log saving or other
+checkpointing mechanisms directly into your application or into your
+`Job` description.
+Job Lifecycle
+When Aurora reads a configuration file and finds a `Job` definition, it:
+1.  Evaluates the `Job` definition.
+2.  Splits the `Job` into its constituent `Task`s.
+3.  Sends those `Task`s to the scheduler.
+4.  The scheduler puts the `Task`s into `PENDING` state, starting each
+    `Task`'s life cycle.
+**Note**: It is not currently possible to create an Aurora job from
+within an Aurora job.
+### Life Of A Task
+![Life of a task](images/lifeofatask.png)
+### PENDING to RUNNING states
+When a `Task` is in the `PENDING` state, the scheduler constantly
+searches for machines satisfying that `Task`'s resource request
+requirements (RAM, disk space, CPU time) while maintaining configuration
+constraints such as "a `Task` must run on machines  dedicated  to a
+particular role" or attribute limit constraints such as "at most 2
+`Task`s from the same `Job` may run on each rack". When the scheduler
+finds a suitable match, it assigns the `Task` to a machine and puts the
+`Task` into the `ASSIGNED` state.
+From the `ASSIGNED` state, the scheduler sends an RPC to the slave
+machine containing `Task` configuration, which the slave uses to spawn
+an executor responsible for the `Task`'s lifecycle. When the scheduler
+receives an acknowledgement that the machine has accepted the `Task`,
+the `Task` goes into `STARTING` state.
+`STARTING` state initializes a `Task` sandbox. When the sandbox is fully
+initialized, Thermos begins to invoke `Process`es. Also, the slave
+machine sends an update to the scheduler that the `Task` is
+in `RUNNING` state.
+If a `Task` stays in `ASSIGNED` or `STARTING` for too long, the
+scheduler forces it into `LOST` state, creating a new `Task` in its
+place that's sent into `PENDING` state. This is technically true of any
+active state: if the Mesos core tells the scheduler that a slave has
+become unhealthy (or outright disappeared), the `Task`s assigned to that
+slave go into `LOST` state and new `Task`s are created in their place.
+From `PENDING` state, there is no guarantee a `Task` will be reassigned
+to the same machine unless job constraints explicitly force it there.
+If there is a state mismatch, (e.g. a machine returns from a `netsplit`
+and the scheduler has marked all its `Task`s `LOST` and rescheduled
+them), a state reconciliation process kills the errant `RUNNING` tasks,
+which may take up to an hour. But to emphasize this point: there is no
+uniqueness guarantee for a single instance of a job in the presence of
+network partitions. If the Task requires that, it should be baked in at
+the application level using a distributed coordination service such as
+### Task Updates
+`Job` configurations can be updated at any point in their lifecycle.
+Usually updates are done incrementally using a process called a *rolling
+upgrade*, in which Tasks are upgraded in small groups, one group at a
+time.  Updates are done using various Aurora Client commands.
+For a configuration update, the Aurora Client calculates required changes
+by examining the current job config state and the new desired job config.
+It then starts a rolling batched update process by going through every batch
+and performing these operations:
+- If an instance is present in the scheduler but isn't in the new config,
+  then that instance is killed.
+- If an instance is not present in the scheduler but is present in
+  the new config, then the instance is created.
+- If an instance is present in both the scheduler the new config, then
+  the client diffs both task configs. If it detects any changes, it
+  performs an instance update by killing the old config instance and adds
+  the new config instance.
+The Aurora client continues through the instance list until all tasks are
+updated, in `RUNNING,` and healthy for a configurable amount of time.
+If the client determines the update is not going well (a percentage of health
+checks have failed), it cancels the update.
+Update cancellation runs a procedure similar to the described above
+update sequence, but in reverse order. New instance configs are swapped
+with old instance configs and batch updates proceed backwards
+from the point where the update failed. E.g.; (0,1,2) (3,4,5) (6,7,
+8-FAIL) results in a rollback in order (8,7,6) (5,4,3) (2,1,0).
+### Giving Priority to Production Tasks: PREEMPTING
+Sometimes a Task needs to be interrupted, such as when a non-production
+Task's resources are needed by a higher priority production Task. This
+type of interruption is called a *pre-emption*. When this happens in
+Aurora, the non-production Task is killed and moved into
+the `PREEMPTING` state  when both the following are true:
+- The task being killed is a non-production task.
+- The other task is a `PENDING` production task that hasn't been
+  scheduled due to a lack of resources.
+Since production tasks are much more important, Aurora kills off the
+non-production task to free up resources for the production task. The
+scheduler UI shows the non-production task was preempted in favor of the
+production task. At some point, tasks in `PREEMPTING` move to `KILLED`.
+Note that non-production tasks consuming many resources are likely to be
+preempted in favor of production tasks.
+### Natural Termination: FINISHED, FAILED
+A `RUNNING` `Task` can terminate without direct user interaction. For
+example, it may be a finite computation that finishes, even something as
+simple as `echo hello world. `Or it could be an exceptional condition in
+a long-lived service. If the `Task` is successful (its underlying
+processes have succeeded with exit status `0` or finished without
+reaching failure limits) it moves into `FINISHED` state. If it finished
+after reaching a set of failure limits, it goes into `FAILED` state.
+### Forceful Termination: KILLING, RESTARTING
+You can terminate a `Task` by issuing an `aurora kill` command, which
+moves it into `KILLING` state. The scheduler then sends the slave  a
+request to terminate the `Task`. If the scheduler receives a successful
+response, it moves the Task into `KILLED` state and never restarts it.
+The scheduler has access to a non-public `RESTARTING` state. If a `Task`
+is forced into the `RESTARTING` state, the scheduler kills the
+underlying task but in parallel schedules an identical replacement for
+You define and configure your Jobs (and their Tasks and Processes) in
+Aurora configuration files. Their filenames end with the `.aurora`
+suffix, and you write them in Python making use of the Pystashio
+templating language, along
+with specific Aurora, Mesos, and Thermos commands and methods. See the
+[Configuration Guide and Reference](/documentation/latest/configuration-reference/) and
+[Configuration Tutorial](/documentation/latest/configuration-tutorial/).
+Creating Jobs
+You create and manipulate Aurora Jobs with the Aurora client, which starts all its
+command line commands with
+`aurora`. See [Aurora Client Commands](/documentation/latest/client-commands/) for details
+about the Aurora Client.
+Interacting With Jobs
+You interact with Aurora jobs either via:
+- Read-only Web UIs
+  Part of the output from creating a new Job is a URL for the Job's scheduler UI page.
+  For example:
+      vagrant@precise64:~$ aurora create example/www-data/prod/hello \
+      /vagrant/examples/jobs/hello_world.aurora
+      INFO] Creating job hello
+      INFO] Response from scheduler: OK (message: 1 new tasks pending for job www-data/prod/hello)
+      INFO] Job url: http://precise64:8081/scheduler/www-data/prod/hello
+  The "Job url" goes to the Job's scheduler UI page. To go to the overall scheduler UI page,
+  stop at the "scheduler" part of the URL, in this case, `http://precise64:8081/scheduler`
+  You can also reach the scheduler UI page via the Client command `aurora open`:
+      aurora open [<cluster>[/<role>[/<env>/<job_name>]]]
+  If only the cluster is specified, it goes directly to that cluster's scheduler main page.
+  If the role is specified, it goes to the top-level role page. If the full job key is specified,
+  it goes directly to the job page where you can inspect individual tasks.
+  Once you click through to a role page, you see Jobs arranged separately by pending jobs, active
+  jobs, and finished jobs. Jobs are arranged by role, typically a service account for production
+  jobs and user accounts for test or development jobs.
+- The Aurora Client's command line interface
+  Several Client commands have a `-o` option that automatically opens a window to
+  the specified Job's scheduler UI URL. And, as described above, the `open` command also takes
+  you there.
+  For a complete list of Aurora Client commands, use `aurora help` and, for specific
+  command help, `aurora help [command]`. **Note**: `aurora help open`
+  returns `"subcommand open not found"` due to our reflection tricks not
+  working on words that are also builtin Python function names. Or see the
+  [Aurora Client Commands](/documentation/latest/client-commands/) document.

Modified: incubator/aurora/site/source/documentation/latest/
--- incubator/aurora/site/source/documentation/latest/ (original)
+++ incubator/aurora/site/source/documentation/latest/ Fri Apr 25 15:59:27 2014
@@ -4,11 +4,11 @@ explore Aurora's various components. To 
 then run `vagrant up` somewhere in the repository source tree to create a team of VMs.  This may take some time initially as it builds all
 the components involved in running an aurora cluster.
-The scheduler is listening on
-The observer is listening on
-The master is listening on
+The scheduler is listening on
+The observer is listening on
+The master is listening on
-Once everything is up, you can `vagrant ssh aurora-scheduler` and execute aurora client commands using the `aurora` client.
+Once everything is up, you can `vagrant ssh devcluster` and execute aurora client commands using the `aurora` client.

View raw message