Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. #11752

anvithks · 2018-12-14T12:12:24Z

Poseidon is the Firmament scheduler integration for Kubernetes. This is the official documentation for Poseidon/Firmament,a new multi-scheduler support for K8S.

As per discussions with @shivramsrivastava @deepak-vij

k8sio-netlify-preview-bot · 2018-12-14T12:12:27Z

Deploy preview for kubernetes-io-vnext-staging processing.

Built with commit c5a72db

https://app.netlify.com/sites/kubernetes-io-vnext-staging/deploys/5c187d7793966b000a69fe2a

shivramsrivastava · 2018-12-14T12:29:18Z

@bsalamat As discussed between you and @deepak-vij , we have raised this PR for adding the official writeup on Poseidon-Firmament.

Please let us know of any changes you require, especially the sub-section under which this doc appears, we have put this under the configuration sub-section, if you want that to be changes please let us know.

bsalamat

Thanks, @anvithks for the PR. I have a few comments. Since this a user doc, we should make it very clear for users where they can benefit from Poseidon and where they should not expect much benefits from it.

bsalamat · 2018-12-14T23:05:25Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+
+{{% capture overview %}}
+
+Poseidon is the Firmament scheduler integration for Kubernetes. At a very high level, Poseidon/Firmament scheduler augments the current Kubernetes scheduling capabilities by incorporating a new novel flow network graph based scheduling capabilities alongside the default Kubernetes Scheduler. It models the scheduling problem as a constraint-based optimization over a flow network graph – by reducing scheduling to a min-cost max-flow optimization problem.


Please provide a link to Firmament.

@bsalamat This is fixed.

bsalamat · 2018-12-14T23:08:27Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+
+Due to the inherent rescheduling capabilities, the new scheduler enables a globally optimal scheduling environment that constantly keeps refining the workloads placements dynamically.
+
+Poseidon/Firmament scheduler runs alongside the default Kubernetes Scheduler as an alternate scheduler – multiple schedulers running simultaneously. As we all know that as part of the Kubernetes multiple schedulers support, each new pod is typically scheduled by the default scheduler, but Kubernetes can be instructed to use another scheduler by specifying the name of another custom scheduler (“Poseidon” in our case) at the time of pod deployment. In this case, the default scheduler will ignore that Pod and allow Poseidon scheduler to schedule the Pod on a relevant node.


Please remove "As we all know that".

Please replace "specifying the name of another custom scheduler (“Poseidon” in our case) at the time of pod deployment" with "specifying the name of another custom scheduler (“Poseidon” in our case) in the PodSpec at the time of pod creation".

@bsalamat This is fixed.

bsalamat · 2018-12-14T23:12:52Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+### Flow graph scheduling based Poseidon/Firmament scheduler provides the following key advantages:  
+- Workloads (pods) are bulk scheduled for enabling scheduling decisions at massive scale.  
+- Based on the extensive performance test results, Poseidon/Firmament scales much better than Kubernetes default scheduler as the number of nodes increase in a cluster. This is due to the fact that Poseidon/Firmament is able to amortize more and more work across workloads.  
+- Poseidon/Firmament Scheduling algorithm outperforms K8S default scheduling algorithm by a wide margin (30X) when it comes to throughput performance numbers for scenarios where compute resource requirements are somewhat uniform across jobs (Replicasets/Deployments/Jobs).  


Please use end to end numbers (including binding time). Users don't care about algorithm performance, they care about rate at which their pods are scheduled. You can point out that, for example, for a 5000-node cluster, Poseidon is more than 4X faster.

@bsalamat This has been updated.

bsalamat · 2018-12-14T23:16:54Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+## Features Comparison Matrix	
+
+
+<table style="border-collapse:collapse;border-spacing:0;border-color:#ccc" class="tg">


Please convert the HTML table to markdown table. There are online tools for doing this. For example: https://jmalarcon.github.io/markdowntables/

@bsalamat This is fixed.

bsalamat · 2018-12-14T23:21:01Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+- Availability of complex rule constraints.  
+- Scheduling in Firmament is very dynamic; it keeps cluster resources in a global optimal state during every scheduling run.  
+- Highly efficient resource utilizations.  
+


Please add a section to note what features of Kubernetes is not yet supported. For example, dynamic volume binding is not supported yet.
Also, point out where Poseidon may be slower than the default scheduler, for example, for pods with inter-pod affinity.

@bsalamat At the moment we have a "Feature Comparison Matrix" table that compares the Default scheduler with Poseidon/Firmament Scheduler. It also has notes with respect to what is currently supported, what is not and what is planned.
This table has been updated now.
Please check and let us now if you feel something is missing.

bsalamat · 2018-12-14T23:23:15Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+
+## Poseidon-Firmament Scheduler - How it works	
+
+As we all know that as part of the Kubernetes multiple schedulers support, each new pod is typically scheduled by the default scheduler, but Kubernetes can be instructed to use another scheduler by specifying the name of another custom scheduler (“Poseidon” in our case) at the time of pod deployment. In this case, the default scheduler will ignore that Pod and allow Poseidon scheduler to schedule the Pod on a relevant node.  


Similar to my comment above, please drop "As we all know" and provide more details on how to set scheduler name in PodSpec.

@bsalamat This is fixed.

bsalamat

Thanks, @anvithks!

bsalamat · 2018-12-17T18:58:29Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+- Workloads (pods) are bulk scheduled for enabling scheduling decisions at massive scale.  
+- Based on the extensive performance test results, Poseidon/Firmament scales much better than Kubernetes default scheduler as the number of nodes increase in a cluster. This is due to the fact that Poseidon/Firmament is able to amortize more and more work across workloads.  
+- Poseidon/Firmament Scheduler outperforms K8S default scheduler by a wide margin when it comes to throughput performance numbers for scenarios where compute resource requirements are somewhat uniform across jobs (Replicasets/Deployments/Jobs). As shown in the graph below, Poseidon/Firmament scheduler end-to-end throughput performance numbers (including bind time) consistently get better and better as the number of nodes in a cluster increase. For example, for a 2,700 nodes cluster (shown in the graph below), Poseidon/Firmament scheduler is 7X (or more) better end-to-end throughput-wise that includes bind time. 
+<br/><br/>


Please remove
s. You can separate paragraphs with a blank line.

@bsalamat Thank you.
This is fixed. Removed the next paragraph so the new line is not required.

bsalamat · 2018-12-17T19:00:49Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+- Based on the extensive performance test results, Poseidon/Firmament scales much better than Kubernetes default scheduler as the number of nodes increase in a cluster. This is due to the fact that Poseidon/Firmament is able to amortize more and more work across workloads.  
+- Poseidon/Firmament Scheduler outperforms K8S default scheduler by a wide margin when it comes to throughput performance numbers for scenarios where compute resource requirements are somewhat uniform across jobs (Replicasets/Deployments/Jobs). As shown in the graph below, Poseidon/Firmament scheduler end-to-end throughput performance numbers (including bind time) consistently get better and better as the number of nodes in a cluster increase. For example, for a 2,700 nodes cluster (shown in the graph below), Poseidon/Firmament scheduler is 7X (or more) better end-to-end throughput-wise that includes bind time. 
+<br/><br/>
+It is also important to highlight that Poseidon/Firmament Scheduling algorithm as it is outperforms K8S default scheduling algorithm by a wide margin (up to 30X or so) when it comes to throughput performance numbers for scenarios where compute resource requirements are somewhat uniform across jobs. In the future, we are planning to further reduce/optimize the bind time by doing bulk bind process in order to make use of bulk scheduling in Poseidon/Firmament in order to realize even better end-to-end throughput performance numbers.    


I think this paragraph does not belong here. This is a user doc, not an academic paper. It should be clear for users. Many users may not now what exactly "end-to-end" and algorithm performance mean. We should provide one set of numbers that are applicable to realistic user scenarios.

@bsalamat This is fixed. Removed the paragraph.

bsalamat · 2018-12-17T19:03:39Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+|Feature|Kubernetes Default Scheduler|Poseidon/Firmament Scheduler|Notes|
+|--- |--- |--- |--- |
+|Node Affinity/Anti-Affinity|Y|Y||
+|Pod Affinity/Anti-Affinity - including support for pod anti-affinity symmetry|Y|Y|Currently, throughput numbers are definitely better for default scheduler versus Poseidon/Firmament due to the recent pod affinity/anti-affinity optimizations within K8S (release 1.11 & 1.12). Poseidon/Firmament scheduler currently “drip-feeds” pods into the scheduling algorithm and processing is one-pod-at-a-time, exactly similar to how default scheduler does. We initially tried achieving pod affinity/anti-affinity functionality using native flow network constructs (convex arc costs, “XOR” & “AND” flow network constructs as described in Ionel Gog’s PHD thesis) but we ran into few challenges and could not make it work.Using the flow network constructs we intend on achieving similar throughput benefits for affinity/anti-affinity pods as we are currently seeing for normal pods.|


Similar comment as above. This paragraph has too much details about the reasons for lower performance of Poseidon. These are not useful in a user doc. I think you should only point out that the default scheduler is faster in such scenarios and your plans to address it in the future.

@bsalamat This is fixed.

bsalamat · 2018-12-17T19:06:12Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+|Node Affinity/Anti-Affinity|Y|Y||
+|Pod Affinity/Anti-Affinity - including support for pod anti-affinity symmetry|Y|Y|Currently, throughput numbers are definitely better for default scheduler versus Poseidon/Firmament due to the recent pod affinity/anti-affinity optimizations within K8S (release 1.11 & 1.12). Poseidon/Firmament scheduler currently “drip-feeds” pods into the scheduling algorithm and processing is one-pod-at-a-time, exactly similar to how default scheduler does. We initially tried achieving pod affinity/anti-affinity functionality using native flow network constructs (convex arc costs, “XOR” & “AND” flow network constructs as described in Ionel Gog’s PHD thesis) but we ran into few challenges and could not make it work.Using the flow network constructs we intend on achieving similar throughput benefits for affinity/anti-affinity pods as we are currently seeing for normal pods.|
+|Taints & Tolerations|Y|Y||
+|Baseline Scheduling capability in accordance to available compute resources (CPU & Memory) on a node|Y|Y**|Not all Predicates & Priorities are supported at this time (although, all the SIG scheduling e2e tests are passing).|


Please remove "all the SIG scheduling e2e tests are passing". End to end tests cover only a small part of the logic which concerns interactions with other components. Most of the functionality of the scheduler is tested in integration and unit-tests.

@bsalamat This is fixed.

bsalamat · 2018-12-17T19:07:57Z

content/en/docs/concepts/configuration/poseidon-firmament-alternate-scheduler.md

+|Taints & Tolerations|Y|Y||
+|Baseline Scheduling capability in accordance to available compute resources (CPU & Memory) on a node|Y|Y**|Not all Predicates & Priorities are supported at this time (although, all the SIG scheduling e2e tests are passing).|
+|Extreme Throughput at scale|Y**|Y|This is due to Poseidon/Firmament bulk scheduling approach superiority versus K8S pod-at-a-time approach. Substantial throughput benefits using Firmament scheduler as long as resource requirements (CPU/Memory) for incoming Pods is uniform across Replicasets/Deployments/Jobs. This is mainly due to efficient amortization of work across Replicasets/Deployments/Jobs . 1) For “Big Data/AI” jobs consisting of large no. of tasks, throughput benefits are tremendous. 2) Substantial throughput benefits also for service or batch job scenarios where workload resource requirements are uniform across Replicasets/Deployments/Jobs.|
+|Optimal Scheduling|Pod-by-Pod scheduler, processes one pod at a time (may result into sub-optimal scheduling)|Bulk Scheduling (Optimal scheduling)|Pod-by-Pod K8S default scheduler may assign tasks to a sub-optimal machine or may end up migrating the first task to another machine – potentially losing all the work the task has done – and replace it with the second task. By contrast, Firmament considers all unscheduled tasks at the same time together with their soft and hard constraints. Thus, avoids unnecessary task migrations and wasting task work.|


The default scheduler never migrates tasks. Please remove that part.

@bsalamat This is fixed. We have removed any reference to task migration.

bsalamat

/lgtm

Thanks, @anvithks!

anvithks · 2018-12-18T05:11:27Z

/assign @MistyHacks

anvithks · 2018-12-19T07:43:18Z

@MistyHacks please do review when you can.

tengqm · 2018-12-23T01:21:57Z

/approve

k8s-ci-robot · 2018-12-23T01:22:25Z

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: tengqm

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

~~OWNERS~~ [tengqm]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

…upport for K8S. (kubernetes#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. (cherry picked from commit 4652684)

…upport for K8S. (kubernetes#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. * Updated the document as per review comments. Added config details. * Updated the document as per the latest review comments. Fixed nits * Made changes as per latest suggestions. * Some more changes added.

…upport for K8S (#12069) * Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. (#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. * Updated the document as per review comments. Added config details. * Updated the document as per the latest review comments. Fixed nits * Made changes as per latest suggestions. * Some more changes added. * Updated as per suggestions. * Changed the release process section. * SIG Docs edits Small edits to match style guidelines. * add plus to feature state * capitalization * revert feature state shortcode since this is a Kubernetes extension, not a direct feature, it shouldn't use the regular feature state tagging.

…upport for K8S (kubernetes#12069) * Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. (kubernetes#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. * Updated the document as per review comments. Added config details. * Updated the document as per the latest review comments. Fixed nits * Made changes as per latest suggestions. * Some more changes added. * Updated as per suggestions. * Changed the release process section. * SIG Docs edits Small edits to match style guidelines. * add plus to feature state * capitalization * revert feature state shortcode since this is a Kubernetes extension, not a direct feature, it shouldn't use the regular feature state tagging. (cherry picked from commit 7730c15)

…12343) * Removed the old version of the Poseidon documentation. Incorrect location. * Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S (#12069) * Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. (#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. * Updated the document as per review comments. Added config details. * Updated the document as per the latest review comments. Fixed nits * Made changes as per latest suggestions. * Some more changes added. * Updated as per suggestions. * Changed the release process section. * SIG Docs edits Small edits to match style guidelines. * add plus to feature state * capitalization * revert feature state shortcode since this is a Kubernetes extension, not a direct feature, it shouldn't use the regular feature state tagging. (cherry picked from commit 7730c15)

…upport for K8S. (kubernetes#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments.

* Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. (#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. * Document timeout attribute for kms-plugin. (#12158) See 72540. * Official documentation on Poseidon/Firmament, a new multi-scheduler (#12343) * Removed the old version of the Poseidon documentation. Incorrect location. * Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S (#12069) * Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. (#11752) * Added documentation about Poseidon-Firmament scheduler * Fixed some style issues. * Udpated the document as per the review comments. * Fixed some typos and updated the document * Updated the document as per the review comments. * Updated the document as per review comments. Added config details. * Updated the document as per the latest review comments. Fixed nits * Made changes as per latest suggestions. * Some more changes added. * Updated as per suggestions. * Changed the release process section. * SIG Docs edits Small edits to match style guidelines. * add plus to feature state * capitalization * revert feature state shortcode since this is a Kubernetes extension, not a direct feature, it shouldn't use the regular feature state tagging. (cherry picked from commit 7730c15) * Remove initializers from doc. It will be removed in 1.14 (#12331) * kubeadm: Document CRI auto detection functionality (#12462) Signed-off-by: Rostislav M. Georgiev <rostislavg@vmware.com> * Minor doc change for GAing Pod DNS Config (#12514) * Graduate ExpandInUsePersistentVolumes feature to beta (#10574) * Rename 2018-11-07-grpc-load-balancing-with-linkerd.md.md file (#12594) * Add dynamic percentage of node scoring to user docs (#12235) * Add dynamic percentage of node scoring to user docs * addressed review comments * delete special symbol (#12445) * Update documentation for VolumeSubpathEnvExpansion (#11843) * Update documentation for VolumeSubpathEnvExpansion * Address comments - improve descriptions * Graduate Pod Priority and Preemption to GA (#12428) * Added Instana links to the documentation (#12977) * Added link to the Instana Kubernetes integration * Added Instana link for services section Added Instana and a link to the Kubernetes integration to the analytics services section and broadened the scope to APM, monitoring and analytics. * Oxford comma /flex * More Oxford commas, because they matter * Update kubectl plugins to stable (#12847) * documentation for CSI topology beta (#12889) * Document changes to default RBAC discovery ClusterRole(Binding)s (#12888) * Document changes to default RBAC discovery ClusterRole(Binding)s Documentation for kubernetes/enhancements#789 and kubernetes/kubernetes#73807 * documentation review feedback * CSI raw block to beta (#12931) * Change incorrect string raw to block (#12926) Fixes #12925 * Update documentation on node OS/arch labels (#12976) These labels have been promoted to GA: kubernetes/enhancements#793 * local pv GA doc updates (#12915) * Publish CRD OpenAPI Documentation (#12910) * add documentation for CustomResourcePublishOpenAPI * address comments fix links, ordered lists, style and typo * kubeadm: add document for upgrading from 1.13 to 1.14 (single CP and HA) (#13189) * kubeadm: add document for upgrading from 1.13 to 1.14 - remove doc for upgrading 1.10 -> 1.11 * kubeadm: apply amends to upgrade-1.14 doc * kubeadm: apply amends to upgrade-1.14 doc (part2) * kubeadm: apply amends to upgrade-1.14 doc (part3) * kubeadm: add note about "upgrade node experimental-control-plane" + add comment about `upgrade plan` * kubeadm: add missing "You should see output similar to this" * fix bullet indentation (#13214) * mark PodReadinessGate GA (#12800) * Update RuntimeClass documentation for beta (#13043) * Update RuntimeClass documentation for beta * Update feature gate & add upgrade section * formatting fixes * Highlight upgrade action required * Address feedback * CSI ephemeral volume alpha documentation (#10934) * update kubectl documentation (#12867) * update kubectl documentation * add document for Secret/ConfigMap generators * replace `kubectl create -f` by `kubectl apply -f` * Add page for kustomization support in kubectl * fix spelling errors and address comments * Documentation for Windows GMSA feature (#12936) * Documentation for Windows GMSA feature Signed-off-by: Deep Debroy <ddebroy@docker.com> * Enhancements to GMSA docs Signed-off-by: Deep Debroy <ddebroy@docker.com> * Fix links Signed-off-by: Deep Debroy <ddebroy@docker.com> * Fix GMSA link Signed-off-by: Deep Debroy <ddebroy@docker.com> * Add GMSA feature flag in feature flag list Signed-off-by: Deep Debroy <ddebroy@docker.com> * Relocate GMSA to container configuration Signed-off-by: Deep Debroy <ddebroy@docker.com> * Add example for container spec Signed-off-by: Deep Debroy <ddebroy@docker.com> * Remove changes in Windows index Signed-off-by: Deep Debroy <ddebroy@docker.com> * Update configure-gmsa.md * Update configure-gmsa.md * Update configure-gmsa.md * Update configure-gmsa.md * Rearrange the steps into two sections and other edits Signed-off-by: Deep Debroy <ddebroy@docker.com> * Fix links Signed-off-by: Deep Debroy <ddebroy@docker.com> * Add reference to script to generate GMSA YAMLs Signed-off-by: Deep Debroy <ddebroy@docker.com> * Some more clarifications for GMSA Signed-off-by: Deep Debroy <ddebroy@docker.com> * HugePages graduated to GA (#13004) * HugePages graduated to GA * fixing nit for build * Docs for node PID limiting (kubernetes/kubernetes#73651) (#12932) * kubeadm: update the reference documentation for 1.14 (#12911) * kubeadm: update list of generated files for 1.14 NOTE: PLACEHOLDERS! these files are generated by SIG Docs each release, but we need them to pass the k/website PR CI. - add join_phase* (new sub phases of join) - add init_phase_upload-certs.md (new upload certs phase for init) - remove alpha-preflight (now both init and join have this) * kubeadm: update reference docs includes for 1.14 - remove includes from alpha.md - add upload-certs to init-phase.md - add join-phase.md and it's phases * kubeadm: update the editorial content of join and init - cleanup master->control-plane node - add some notes about phases and join - remove table about pre-pulling images - remove outdated info about self-hosting * kubeadm: update target release for v1alpha3 removal 1.14 -> 1.15 * kubeadm: copy edits for 1.14 reference docs (part1) * kubeadm: use "shell" for code blocks * kubeadm: update the 1.14 HA guide (#13191) * kubeadm: update the 1.14 HA guide * kubeadm: try to fix note/caution indent in HA page * kubeadm: fix missing sudo and minor amends in HA doc * kubeadm: apply latest amends to the HA doc for 1.14 * fixed a few missed merge conflicts * Admission Webhook new features doc (#12938) - kubernetes/kubernetes#74998 - kubernetes/kubernetes#74477 - kubernetes/kubernetes#74562 * Clarifications and fixes in GMSA doc (#13226) * Clarifications and fixes in GMSA doc Signed-off-by: Deep Debroy <ddebroy@docker.com> * Update configure-gmsa.md * Reformat to align headings and pre-reqs better Signed-off-by: Deep Debroy <ddebroy@docker.com> * Reformat to align headings and pre-reqs better Signed-off-by: Deep Debroy <ddebroy@docker.com> * Reformat to fix bullets Signed-off-by: Deep Debroy <ddebroy@docker.com> * Reword application of sample gmsa Signed-off-by: Deep Debroy <ddebroy@docker.com> * Update configure-gmsa.md * Address feedback to use active voice Signed-off-by: Deep Debroy <ddebroy@docker.com> * Address feedback to use active voice Signed-off-by: Deep Debroy <ddebroy@docker.com> * RunAsGroup documentation for Progressing this to Beta (#12297) * start serverside-apply documentation (#13077) * start serverside-apply documentation * add more concept info on server side apply * Update api concepts * Update api-concepts.md * fix style issues * Document CSI update (#12928) * Document CSI update * Finish CSI documentation Also fix mistake with ExpandInUsePersistentVolumes documented as beta * Overall docs for CSI Migration feature (#12935) * Placeholder docs for CSI Migration feature Signed-off-by: Deep Debroy <ddebroy@docker.com> * Address CR comments and update feature gates Signed-off-by: Deep Debroy <ddebroy@docker.com> * Add mappings for CSI plugins Signed-off-by: Deep Debroy <ddebroy@docker.com> * Add sections for AWS and GCE PD migration Signed-off-by: Deep Debroy <ddebroy@docker.com> * Add docs for Cinder and CSI Migration info Signed-off-by: Deep Debroy <ddebroy@docker.com> * Clarify scope to volumes with file system Signed-off-by: Deep Debroy <ddebroy@docker.com> * Change the format of EBS and Cinder CSI Migration sections to follow the GCE template Signed-off-by: Deep Debroy <ddebroy@docker.com> * Windows documentation updates for 1.14 (#12929) * Updated the note to indicate doc work for 1.14 * first attempt at md export from gdoc * simplifyig * big attempt * moving DRAFT windows content to PR for review * moving content to PR in markdown for review * updated note tags * Delete windows-contributing.md deleting this file as it is already ported to the github contributor guide * fixed formatting in intro and cluster setup guide * updating formatting for running containers guide * rejiggered end of troubleshooting * fixed minor typos * Clarified the windows binary download step * Update _index.md making updates based on feedback * Update _index.md updating ovn-kubernetes docs * Update _index.md * Update _index.md * updating relative docs links updating all the links to be relative links to /docs * Update _index.md * Update _index.md updates for windows services and ovn-kubernetes * formatted for correct step numbering * fix typos * Update _index.md updates for flannel PR in troubleshooting * Update _index.md * Update _index.md updating a few sections like roadmap, services, troubleshooting/filing tickets * Update _index.md * Update _index.md * Update _index.md * Fixed a few whitespace issues * Update _index.md * Update _index.md * Update _index.md * add section on upgrading CoreDNS (#12909) * documentation for kubelet resource metrics endpoint (#12934) * windows docs updates for 1.14 (#13279) * Delete sample-l2bridge-wincni-config.json this file is not used anywhere * Update _index.md * Update _index.md * Update _index.md * Update _index.md * Update _index.md * Rename content/en/docs/getting-started-guides/windows/_index.md to content/en/docs/setup/windows/_index.md moving to new location * Delete flannel-master-kubectl-get-ds.png * Delete flannel-master-kubeclt-get-pods.png * Delete windows-docker-error.png * Add files via upload * Rename _index.md to add-windows-nodes.md * Create _index.md * Update _index.md * Update add-windows-nodes.md * Update add-windows-nodes.md * Create user-guide-windows-nodes.md * Create user-guide-windows-containers.md * Update and rename add-windows-nodes.md to intro-windows-nodes.md * Update user-guide-windows-containers.md * Rename intro-windows-nodes.md to intro-windows-in-kubernetes.md * Update user-guide-windows-nodes.md * Update user-guide-windows-containers.md * Update user-guide-windows-containers.md * Update user-guide-windows-nodes.md * Update user-guide-windows-containers.md * Update _index.md * Update intro-windows-in-kubernetes.md * Update intro-windows-in-kubernetes.md fixing the pause image * Update intro-windows-in-kubernetes.md changing tables from html to MD * Update user-guide-windows-nodes.md converting tables from HTML to MD * Update intro-windows-in-kubernetes.md * Update user-guide-windows-nodes.md * Update user-guide-windows-nodes.md * Update user-guide-windows-nodes.md updating the numbering , even though it messes up the notes a little bit. Jim will file a ticket to follow up * Update user-guide-windows-nodes.md * update to windows docs for 1.14 (#13322) * Update intro-windows-in-kubernetes.md * Update intro-windows-in-kubernetes.md * Update intro-windows-in-kubernetes.md * Update intro-windows-in-kubernetes.md * Update intro-windows-in-kubernetes.md * Update user-guide-windows-containers.md * Update user-guide-windows-nodes.md * Update intro-windows-in-kubernetes.md (#13344) * server side apply followup (#13321) * change some parts of serverside apply docs in response to comments * fix typos and wording * Update config.toml (#13365)

Added documentation about Poseidon-Firmament scheduler

8a3c35d

k8s-ci-robot added the cncf-cla: yes Indicates the PR's author has signed the CNCF CLA. label Dec 14, 2018

k8s-ci-robot added size/L Denotes a PR that changes 100-499 lines, ignoring generated files. language/en Issues or PRs related to English language labels Dec 14, 2018

k8s-ci-robot requested review from jimangel and xiangpengzhao December 14, 2018 12:12

Fixed some style issues.

c80a5e3

bsalamat reviewed Dec 14, 2018

View reviewed changes

anvithks added 2 commits December 17, 2018 09:17

Udpated the document as per the review comments.

d0e8f73

Fixed some typos and updated the document

9aa2fa9

bsalamat reviewed Dec 17, 2018

View reviewed changes

Updated the document as per the review comments.

c5a72db

k8s-ci-robot assigned bsalamat Dec 18, 2018

bsalamat reviewed Dec 18, 2018

View reviewed changes

k8s-ci-robot added the lgtm "Looks good to me", indicates that a PR is ready to be merged. label Dec 18, 2018

k8s-ci-robot assigned mdlinville Dec 18, 2018

k8s-ci-robot added the approved Indicates a PR has been approved by an approver from all required OWNERS files. label Dec 23, 2018

k8s-ci-robot merged commit 4652684 into kubernetes:dev-1.14 Dec 23, 2018

anvithks mentioned this pull request Jan 4, 2019

Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S #12069

Merged

anvithks mentioned this pull request Jan 23, 2019

Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S #12340

Closed

anvithks mentioned this pull request Jan 23, 2019

Official documentation on Poseidon/Firmament, a new multi-scheduler #12343

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. #11752

Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. #11752

anvithks commented Dec 14, 2018

k8sio-netlify-preview-bot commented Dec 14, 2018 •

edited

Loading

shivramsrivastava commented Dec 14, 2018

bsalamat left a comment

bsalamat Dec 14, 2018

anvithks Dec 17, 2018

bsalamat Dec 14, 2018

anvithks Dec 17, 2018

bsalamat Dec 14, 2018

anvithks Dec 17, 2018

bsalamat Dec 14, 2018

anvithks Dec 17, 2018

bsalamat Dec 14, 2018

anvithks Dec 17, 2018

bsalamat Dec 14, 2018

anvithks Dec 17, 2018

bsalamat left a comment

bsalamat Dec 17, 2018

anvithks Dec 18, 2018

bsalamat Dec 17, 2018

anvithks Dec 18, 2018

bsalamat Dec 17, 2018

anvithks Dec 18, 2018

bsalamat Dec 17, 2018

anvithks Dec 18, 2018

bsalamat Dec 17, 2018

anvithks Dec 18, 2018

bsalamat left a comment

anvithks commented Dec 18, 2018

anvithks commented Dec 19, 2018

tengqm commented Dec 23, 2018

k8s-ci-robot commented Dec 23, 2018


		{{% capture overview %}}

		Poseidon is the Firmament scheduler integration for Kubernetes. At a very high level, Poseidon/Firmament scheduler augments the current Kubernetes scheduling capabilities by incorporating a new novel flow network graph based scheduling capabilities alongside the default Kubernetes Scheduler. It models the scheduling problem as a constraint-based optimization over a flow network graph – by reducing scheduling to a min-cost max-flow optimization problem.


		Due to the inherent rescheduling capabilities, the new scheduler enables a globally optimal scheduling environment that constantly keeps refining the workloads placements dynamically.

		Poseidon/Firmament scheduler runs alongside the default Kubernetes Scheduler as an alternate scheduler – multiple schedulers running simultaneously. As we all know that as part of the Kubernetes multiple schedulers support, each new pod is typically scheduled by the default scheduler, but Kubernetes can be instructed to use another scheduler by specifying the name of another custom scheduler (“Poseidon” in our case) at the time of pod deployment. In this case, the default scheduler will ignore that Pod and allow Poseidon scheduler to schedule the Pod on a relevant node.

		## Features Comparison Matrix


		<table style="border-collapse:collapse;border-spacing:0;border-color:#ccc" class="tg">


		## Poseidon-Firmament Scheduler - How it works

		As we all know that as part of the Kubernetes multiple schedulers support, each new pod is typically scheduled by the default scheduler, but Kubernetes can be instructed to use another scheduler by specifying the name of another custom scheduler (“Poseidon” in our case) at the time of pod deployment. In this case, the default scheduler will ignore that Pod and allow Poseidon scheduler to schedule the Pod on a relevant node.

Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. #11752

Official documentation on Poseidon/Firmament, a new multi-scheduler support for K8S. #11752

Conversation

anvithks commented Dec 14, 2018

k8sio-netlify-preview-bot commented Dec 14, 2018 • edited Loading

shivramsrivastava commented Dec 14, 2018

bsalamat left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bsalamat left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

bsalamat left a comment

Choose a reason for hiding this comment

anvithks commented Dec 18, 2018

anvithks commented Dec 19, 2018

tengqm commented Dec 23, 2018

k8s-ci-robot commented Dec 23, 2018

k8sio-netlify-preview-bot commented Dec 14, 2018 •

edited

Loading