Changes in docs for release: v0.9.0

Author: codeflare-machine-account

docs/cluster/cluster.html

@@ -141,6 +141,9 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.cluster</code></h1>
 
         name = self.config.name
         namespace = self.config.namespace
+        head_cpus = self.config.head_cpus
+        head_memory = self.config.head_memory
+        head_gpus = self.config.head_gpus
         min_cpu = self.config.min_cpus
         max_cpu = self.config.max_cpus
         min_memory = self.config.min_memory
@@ -158,6 +161,9 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.cluster</code></h1>
         return generate_appwrapper(
             name=name,
             namespace=namespace,
+            head_cpus=head_cpus,
+            head_memory=head_memory,
+            head_gpus=head_gpus,
             min_cpu=min_cpu,
             max_cpu=max_cpu,
             min_memory=min_memory,
@@ -290,7 +296,7 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.cluster</code></h1>
         else:
             return False
 
-    def wait_ready(self, timeout: Optional[int] = None):
+    def wait_ready(self, timeout: Optional[int] = None, dashboard_check: bool = True):
         &#34;&#34;&#34;
         Waits for requested cluster to be ready, up to an optional timeout (s).
         Checks every five seconds.
@@ -300,19 +306,32 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.cluster</code></h1>
         dashboard_ready = False
         status = None
         time = 0
-        while not ready or not dashboard_ready:
+        while not ready:
             status, ready = self.status(print_to_console=False)
-            dashboard_ready = self.is_dashboard_ready()
             if status == CodeFlareClusterStatus.UNKNOWN:
                 print(
                     &#34;WARNING: Current cluster status is unknown, have you run cluster.up yet?&#34;
                 )
-            if not ready or not dashboard_ready:
+            if not ready:
+                if timeout and time &gt;= timeout:
+                    raise TimeoutError(
+                        f&#34;wait() timed out after waiting {timeout}s for cluster to be ready&#34;
+                    )
+                sleep(5)
+                time += 5
+        print(&#34;Requested cluster is up and running!&#34;)
+
+        while dashboard_check and not dashboard_ready:
+            dashboard_ready = self.is_dashboard_ready()
+            if not dashboard_ready:
                 if timeout and time &gt;= timeout:
-                    raise TimeoutError(f&#34;wait() timed out after waiting {timeout}s&#34;)
+                    raise TimeoutError(
+                        f&#34;wait() timed out after waiting {timeout}s for dashboard to be ready&#34;
+                    )
                 sleep(5)
                 time += 5
-        print(&#34;Requested cluster and dashboard are up and running!&#34;)
+        if dashboard_ready:
+            print(&#34;Dashboard is ready!&#34;)
 
     def details(self, print_to_console: bool = True) -&gt; RayCluster:
         cluster = _copy_to_ray(self)
@@ -640,6 +659,15 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.cluster</code></h1>
         worker_gpu=0,  # hard to detect currently how many gpus, can override it with what the user asked for
         namespace=rc[&#34;metadata&#34;][&#34;namespace&#34;],
         dashboard=ray_route,
+        head_cpus=rc[&#34;spec&#34;][&#34;headGroupSpec&#34;][&#34;template&#34;][&#34;spec&#34;][&#34;containers&#34;][0][
+            &#34;resources&#34;
+        ][&#34;limits&#34;][&#34;cpu&#34;],
+        head_mem=rc[&#34;spec&#34;][&#34;headGroupSpec&#34;][&#34;template&#34;][&#34;spec&#34;][&#34;containers&#34;][0][
+            &#34;resources&#34;
+        ][&#34;limits&#34;][&#34;memory&#34;],
+        head_gpu=rc[&#34;spec&#34;][&#34;headGroupSpec&#34;][&#34;template&#34;][&#34;spec&#34;][&#34;containers&#34;][0][
+            &#34;resources&#34;
+        ][&#34;limits&#34;][&#34;nvidia.com/gpu&#34;],
     )
 
 
@@ -670,6 +698,9 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.cluster</code></h1>
         worker_gpu=cluster.config.num_gpus,
         namespace=cluster.config.namespace,
         dashboard=cluster.cluster_dashboard_uri(),
+        head_cpus=cluster.config.head_cpus,
+        head_mem=cluster.config.head_memory,
+        head_gpu=cluster.config.head_gpus,
     )
     if ray.status == CodeFlareClusterStatus.READY:
         ray.status = RayClusterStatus.READY
@@ -879,6 +910,9 @@ <h2 class="section-title" id="header-classes">Classes</h2>
 
         name = self.config.name
         namespace = self.config.namespace
+        head_cpus = self.config.head_cpus
+        head_memory = self.config.head_memory
+        head_gpus = self.config.head_gpus
         min_cpu = self.config.min_cpus
         max_cpu = self.config.max_cpus
         min_memory = self.config.min_memory
@@ -896,6 +930,9 @@ <h2 class="section-title" id="header-classes">Classes</h2>
         return generate_appwrapper(
             name=name,
             namespace=namespace,
+            head_cpus=head_cpus,
+            head_memory=head_memory,
+            head_gpus=head_gpus,
             min_cpu=min_cpu,
             max_cpu=max_cpu,
             min_memory=min_memory,
@@ -1028,7 +1065,7 @@ <h2 class="section-title" id="header-classes">Classes</h2>
         else:
             return False
 
-    def wait_ready(self, timeout: Optional[int] = None):
+    def wait_ready(self, timeout: Optional[int] = None, dashboard_check: bool = True):
         &#34;&#34;&#34;
         Waits for requested cluster to be ready, up to an optional timeout (s).
         Checks every five seconds.
@@ -1038,19 +1075,32 @@ <h2 class="section-title" id="header-classes">Classes</h2>
         dashboard_ready = False
         status = None
         time = 0
-        while not ready or not dashboard_ready:
+        while not ready:
             status, ready = self.status(print_to_console=False)
-            dashboard_ready = self.is_dashboard_ready()
             if status == CodeFlareClusterStatus.UNKNOWN:
                 print(
                     &#34;WARNING: Current cluster status is unknown, have you run cluster.up yet?&#34;
                 )
-            if not ready or not dashboard_ready:
+            if not ready:
+                if timeout and time &gt;= timeout:
+                    raise TimeoutError(
+                        f&#34;wait() timed out after waiting {timeout}s for cluster to be ready&#34;
+                    )
+                sleep(5)
+                time += 5
+        print(&#34;Requested cluster is up and running!&#34;)
+
+        while dashboard_check and not dashboard_ready:
+            dashboard_ready = self.is_dashboard_ready()
+            if not dashboard_ready:
                 if timeout and time &gt;= timeout:
-                    raise TimeoutError(f&#34;wait() timed out after waiting {timeout}s&#34;)
+                    raise TimeoutError(
+                        f&#34;wait() timed out after waiting {timeout}s for dashboard to be ready&#34;
+                    )
                 sleep(5)
                 time += 5
-        print(&#34;Requested cluster and dashboard are up and running!&#34;)
+        if dashboard_ready:
+            print(&#34;Dashboard is ready!&#34;)
 
     def details(self, print_to_console: bool = True) -&gt; RayCluster:
         cluster = _copy_to_ray(self)
@@ -1267,6 +1317,9 @@ <h3>Methods</h3>
 
     name = self.config.name
     namespace = self.config.namespace
+    head_cpus = self.config.head_cpus
+    head_memory = self.config.head_memory
+    head_gpus = self.config.head_gpus
     min_cpu = self.config.min_cpus
     max_cpu = self.config.max_cpus
     min_memory = self.config.min_memory
@@ -1284,6 +1337,9 @@ <h3>Methods</h3>
     return generate_appwrapper(
         name=name,
         namespace=namespace,
+        head_cpus=head_cpus,
+        head_memory=head_memory,
+        head_gpus=head_gpus,
         min_cpu=min_cpu,
         max_cpu=max_cpu,
         min_memory=min_memory,
@@ -1653,7 +1709,7 @@ <h3>Methods</h3>
 </details>
 </dd>
 <dt id="codeflare_sdk.cluster.cluster.Cluster.wait_ready"><code class="name flex">
-<span>def <span class="ident">wait_ready</span></span>(<span>self, timeout: Optional[int] = None)</span>
+<span>def <span class="ident">wait_ready</span></span>(<span>self, timeout: Optional[int] = None, dashboard_check: bool = True)</span>
 </code></dt>
 <dd>
 <div class="desc"><p>Waits for requested cluster to be ready, up to an optional timeout (s).
@@ -1662,7 +1718,7 @@ <h3>Methods</h3>
 <summary>
 <span>Expand source code</span>
 </summary>
-<pre><code class="python">def wait_ready(self, timeout: Optional[int] = None):
+<pre><code class="python">def wait_ready(self, timeout: Optional[int] = None, dashboard_check: bool = True):
     &#34;&#34;&#34;
     Waits for requested cluster to be ready, up to an optional timeout (s).
     Checks every five seconds.
@@ -1672,19 +1728,32 @@ <h3>Methods</h3>
     dashboard_ready = False
     status = None
     time = 0
-    while not ready or not dashboard_ready:
+    while not ready:
         status, ready = self.status(print_to_console=False)
-        dashboard_ready = self.is_dashboard_ready()
         if status == CodeFlareClusterStatus.UNKNOWN:
             print(
                 &#34;WARNING: Current cluster status is unknown, have you run cluster.up yet?&#34;
             )
-        if not ready or not dashboard_ready:
+        if not ready:
+            if timeout and time &gt;= timeout:
+                raise TimeoutError(
+                    f&#34;wait() timed out after waiting {timeout}s for cluster to be ready&#34;
+                )
+            sleep(5)
+            time += 5
+    print(&#34;Requested cluster is up and running!&#34;)
+
+    while dashboard_check and not dashboard_ready:
+        dashboard_ready = self.is_dashboard_ready()
+        if not dashboard_ready:
             if timeout and time &gt;= timeout:
-                raise TimeoutError(f&#34;wait() timed out after waiting {timeout}s&#34;)
+                raise TimeoutError(
+                    f&#34;wait() timed out after waiting {timeout}s for dashboard to be ready&#34;
+                )
             sleep(5)
             time += 5
-    print(&#34;Requested cluster and dashboard are up and running!&#34;)</code></pre>
+    if dashboard_ready:
+        print(&#34;Dashboard is ready!&#34;)</code></pre>
 </details>
 </dd>
 </dl>

docs/cluster/config.html

@@ -66,6 +66,9 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.config</code></h1>
     name: str
     namespace: str = None
     head_info: list = field(default_factory=list)
+    head_cpus: int = 2
+    head_memory: int = 8
+    head_gpus: int = 0
     machine_types: list = field(default_factory=list)  # [&#34;m4.xlarge&#34;, &#34;g4dn.xlarge&#34;]
     min_cpus: int = 1
     max_cpus: int = 1
@@ -76,7 +79,7 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.config</code></h1>
     template: str = f&#34;{dir}/templates/base-template.yaml&#34;
     instascale: bool = False
     envs: dict = field(default_factory=dict)
-    image: str = &#34;quay.io/project-codeflare/ray:2.5.0-py38-cu116&#34;
+    image: str = &#34;quay.io/project-codeflare/ray:latest-py39-cu118&#34;
     local_interactive: bool = False
     image_pull_secrets: list = field(default_factory=list)
     dispatch_priority: str = None</code></pre>
@@ -93,7 +96,7 @@ <h2 class="section-title" id="header-classes">Classes</h2>
 <dl>
 <dt id="codeflare_sdk.cluster.config.ClusterConfiguration"><code class="flex name class">
 <span>class <span class="ident">ClusterConfiguration</span></span>
-<span>(</span><span>name: str, namespace: str = None, head_info: list = &lt;factory&gt;, machine_types: list = &lt;factory&gt;, min_cpus: int = 1, max_cpus: int = 1, num_workers: int = 1, min_memory: int = 2, max_memory: int = 2, num_gpus: int = 0, template: str = '/home/runner/work/codeflare-sdk/codeflare-sdk/src/codeflare_sdk/templates/base-template.yaml', instascale: bool = False, envs: dict = &lt;factory&gt;, image: str = 'quay.io/project-codeflare/ray:2.5.0-py38-cu116', local_interactive: bool = False, image_pull_secrets: list = &lt;factory&gt;, dispatch_priority: str = None)</span>
+<span>(</span><span>name: str, namespace: str = None, head_info: list = &lt;factory&gt;, head_cpus: int = 2, head_memory: int = 8, head_gpus: int = 0, machine_types: list = &lt;factory&gt;, min_cpus: int = 1, max_cpus: int = 1, num_workers: int = 1, min_memory: int = 2, max_memory: int = 2, num_gpus: int = 0, template: str = '/home/runner/work/codeflare-sdk/codeflare-sdk/src/codeflare_sdk/templates/base-template.yaml', instascale: bool = False, envs: dict = &lt;factory&gt;, image: str = 'quay.io/project-codeflare/ray:latest-py39-cu118', local_interactive: bool = False, image_pull_secrets: list = &lt;factory&gt;, dispatch_priority: str = None)</span>
 </code></dt>
 <dd>
 <div class="desc"><p>This dataclass is used to specify resource requirements and other details, and
@@ -111,6 +114,9 @@ <h2 class="section-title" id="header-classes">Classes</h2>
     name: str
     namespace: str = None
     head_info: list = field(default_factory=list)
+    head_cpus: int = 2
+    head_memory: int = 8
+    head_gpus: int = 0
     machine_types: list = field(default_factory=list)  # [&#34;m4.xlarge&#34;, &#34;g4dn.xlarge&#34;]
     min_cpus: int = 1
     max_cpus: int = 1
@@ -121,7 +127,7 @@ <h2 class="section-title" id="header-classes">Classes</h2>
     template: str = f&#34;{dir}/templates/base-template.yaml&#34;
     instascale: bool = False
     envs: dict = field(default_factory=dict)
-    image: str = &#34;quay.io/project-codeflare/ray:2.5.0-py38-cu116&#34;
+    image: str = &#34;quay.io/project-codeflare/ray:latest-py39-cu118&#34;
     local_interactive: bool = False
     image_pull_secrets: list = field(default_factory=list)
     dispatch_priority: str = None</code></pre>
@@ -136,10 +142,22 @@ <h3>Class variables</h3>
 <dd>
 <div class="desc"></div>
 </dd>
+<dt id="codeflare_sdk.cluster.config.ClusterConfiguration.head_cpus"><code class="name">var <span class="ident">head_cpus</span> : int</code></dt>
+<dd>
+<div class="desc"></div>
+</dd>
+<dt id="codeflare_sdk.cluster.config.ClusterConfiguration.head_gpus"><code class="name">var <span class="ident">head_gpus</span> : int</code></dt>
+<dd>
+<div class="desc"></div>
+</dd>
 <dt id="codeflare_sdk.cluster.config.ClusterConfiguration.head_info"><code class="name">var <span class="ident">head_info</span> : list</code></dt>
 <dd>
 <div class="desc"></div>
 </dd>
+<dt id="codeflare_sdk.cluster.config.ClusterConfiguration.head_memory"><code class="name">var <span class="ident">head_memory</span> : int</code></dt>
+<dd>
+<div class="desc"></div>
+</dd>
 <dt id="codeflare_sdk.cluster.config.ClusterConfiguration.image"><code class="name">var <span class="ident">image</span> : str</code></dt>
 <dd>
 <div class="desc"></div>
@@ -219,7 +237,10 @@ <h4><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration" href="#co
 <ul class="two-column">
 <li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.dispatch_priority" href="#codeflare_sdk.cluster.config.ClusterConfiguration.dispatch_priority">dispatch_priority</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.envs" href="#codeflare_sdk.cluster.config.ClusterConfiguration.envs">envs</a></code></li>
+<li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.head_cpus" href="#codeflare_sdk.cluster.config.ClusterConfiguration.head_cpus">head_cpus</a></code></li>
+<li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.head_gpus" href="#codeflare_sdk.cluster.config.ClusterConfiguration.head_gpus">head_gpus</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.head_info" href="#codeflare_sdk.cluster.config.ClusterConfiguration.head_info">head_info</a></code></li>
+<li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.head_memory" href="#codeflare_sdk.cluster.config.ClusterConfiguration.head_memory">head_memory</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.image" href="#codeflare_sdk.cluster.config.ClusterConfiguration.image">image</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.image_pull_secrets" href="#codeflare_sdk.cluster.config.ClusterConfiguration.image_pull_secrets">image_pull_secrets</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.config.ClusterConfiguration.instascale" href="#codeflare_sdk.cluster.config.ClusterConfiguration.instascale">instascale</a></code></li>

docs/cluster/model.html

@@ -102,6 +102,9 @@ <h1 class="title">Module <code>codeflare_sdk.cluster.model</code></h1>
 
     name: str
     status: RayClusterStatus
+    head_cpus: int
+    head_mem: str
+    head_gpu: int
     workers: int
     worker_mem_min: str
     worker_mem_max: str
@@ -287,7 +290,7 @@ <h3>Class variables</h3>
 </dd>
 <dt id="codeflare_sdk.cluster.model.RayCluster"><code class="flex name class">
 <span>class <span class="ident">RayCluster</span></span>
-<span>(</span><span>name: str, status: <a title="codeflare_sdk.cluster.model.RayClusterStatus" href="#codeflare_sdk.cluster.model.RayClusterStatus">RayClusterStatus</a>, workers: int, worker_mem_min: str, worker_mem_max: str, worker_cpu: int, worker_gpu: int, namespace: str, dashboard: str)</span>
+<span>(</span><span>name: str, status: <a title="codeflare_sdk.cluster.model.RayClusterStatus" href="#codeflare_sdk.cluster.model.RayClusterStatus">RayClusterStatus</a>, head_cpus: int, head_mem: str, head_gpu: int, workers: int, worker_mem_min: str, worker_mem_max: str, worker_cpu: int, worker_gpu: int, namespace: str, dashboard: str)</span>
 </code></dt>
 <dd>
 <div class="desc"><p>For storing information about a Ray cluster.</p></div>
@@ -302,6 +305,9 @@ <h3>Class variables</h3>
 
     name: str
     status: RayClusterStatus
+    head_cpus: int
+    head_mem: str
+    head_gpu: int
     workers: int
     worker_mem_min: str
     worker_mem_max: str
@@ -316,6 +322,18 @@ <h3>Class variables</h3>
 <dd>
 <div class="desc"></div>
 </dd>
+<dt id="codeflare_sdk.cluster.model.RayCluster.head_cpus"><code class="name">var <span class="ident">head_cpus</span> : int</code></dt>
+<dd>
+<div class="desc"></div>
+</dd>
+<dt id="codeflare_sdk.cluster.model.RayCluster.head_gpu"><code class="name">var <span class="ident">head_gpu</span> : int</code></dt>
+<dd>
+<div class="desc"></div>
+</dd>
+<dt id="codeflare_sdk.cluster.model.RayCluster.head_mem"><code class="name">var <span class="ident">head_mem</span> : str</code></dt>
+<dd>
+<div class="desc"></div>
+</dd>
 <dt id="codeflare_sdk.cluster.model.RayCluster.name"><code class="name">var <span class="ident">name</span> : str</code></dt>
 <dd>
 <div class="desc"></div>
@@ -447,6 +465,9 @@ <h4><code><a title="codeflare_sdk.cluster.model.CodeFlareClusterStatus" href="#c
 <h4><code><a title="codeflare_sdk.cluster.model.RayCluster" href="#codeflare_sdk.cluster.model.RayCluster">RayCluster</a></code></h4>
 <ul class="two-column">
 <li><code><a title="codeflare_sdk.cluster.model.RayCluster.dashboard" href="#codeflare_sdk.cluster.model.RayCluster.dashboard">dashboard</a></code></li>
+<li><code><a title="codeflare_sdk.cluster.model.RayCluster.head_cpus" href="#codeflare_sdk.cluster.model.RayCluster.head_cpus">head_cpus</a></code></li>
+<li><code><a title="codeflare_sdk.cluster.model.RayCluster.head_gpu" href="#codeflare_sdk.cluster.model.RayCluster.head_gpu">head_gpu</a></code></li>
+<li><code><a title="codeflare_sdk.cluster.model.RayCluster.head_mem" href="#codeflare_sdk.cluster.model.RayCluster.head_mem">head_mem</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.model.RayCluster.name" href="#codeflare_sdk.cluster.model.RayCluster.name">name</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.model.RayCluster.namespace" href="#codeflare_sdk.cluster.model.RayCluster.namespace">namespace</a></code></li>
 <li><code><a title="codeflare_sdk.cluster.model.RayCluster.status" href="#codeflare_sdk.cluster.model.RayCluster.status">status</a></code></li>