Notes:

  • Priti Desai is moderating today
  • last meeting was 2018-05-09
  • Priti, Rob, Dave G., Sam, Matt, Olivier, Neeraj, Dominic, Andy,Tzu Chiao, Tyson, Justin, Chetan, Himavanth, Michele, Carlos, Sandeep, Justin, Vincent

Introductions of new attendees

  • Dominic: working in Korea on their platform using OW, time is not the best to attend in-person (midnight) excited to join!
  • Sam Baxter - could not speak, maybe next call
  • Tzu - Chiao - could not speak, maybe next call


Open comments on status/updates in a few areas:

  • Main/core OpenWhisk
    • https://github.com/apache/incubator-openwhisk/pulls?utf8=%E2%9C%93&q=is%3Apr+is%3Amerged
    • Status updates:
      • Carlos: James made updates to swift runtime (version updates)
      • Docker skeleton updated SSH client to alpine based on user feedback, fixed working with various SSH clients
      • Main README updated to point to docker compose; Docker Mac and Linux is faster to deploy with Docker Compose, less control than Ansible (alternative to Vagrant)
      • Chance from Adobe helped with Arch. diagram to remove Consul 9suing Draw.io)
      • James pointed out email a corner case (Eric and Carlos worked on) some swift not compiling ; notified Swift compiler team, some updates made to runtime to “workaround” (optimization flag had to be turned off). Note updated on IBM cloud too.
      • Vagrant on Window was broken (soft links don't work), so again now use CLI remote.
      • Nadeem working on nginx security improvements
      • Rodric helping improve Gradle tests and Devops (cleanup); cleaner log messages too.
      • Dave showing an IBM cluster as well, it works nicely with simple changes to make Ingress work there (believe this proves it can work with any vendors based upon this test). YAML files on command line can help configure the needed ingress (etc.)
      • Dave: Please help add instructions to Docs to setup in other Clouds!
    • Scheduling Algo. change proposal:
      • Dominic: (share presentation)
        • Issues: Hash is used to choose HomeInvoker
          • capacity of Invokers is NOT considered
          • Slow Docker commands not considered
            • Docker pause slow for example
          • Container deletion… takes 700+ ms.
        • Scenario: Action A & B both have hash 0 (1 controller)
          • both will be sched. to Invoker 0
          • BUT Invokers 1 & 2 are not used
          • Action A & B now are slowed (because all going to Invoker 0)
        • Other scenarios: forces “cold” starts to occur; removal of Warm containers is perhaps not efficient
        • Note: Sometimes waiting for Invoker to complete a previous Action may be faster than waiting for a Cold start
        • Note: Sometimes even when # of Concurrent running containers is less than limit, controller always returns “too many requests” until it receives completion message form Invokers
        • Summary: Nondeterministic performance (TPS)
          • Do not know when we need to add more servers
          • TPS is no proportional to # of invokers
          • TPS is highly dep. on container creation/deletion
        • New Sched. Algo:
          • Note: Many things not complete or to be defined…
          • Want more people involved in discussion
        • Priti: may need to resume next meeting
        • Matt: please upload and provide link (Cwiki), email Carlos or myself if you need us to post for you
        • Carlos: using topics may not allow us to use this/scale
    • Go Runtime / Docker Skeleton (handlers for perf. improvements based upon Go experience) (Michele/Carlos)
      • No update
  • Release process:
    • Matt: reviews license compliance document pages (inclusion/exclusion tables) and shows table of all files that do nothave headers.
      • Target next week (Wed.) to start source release process
    • Carlos: do not include compiled artifacts do not need to be included
    • Matt: agrees but will get opinion on mailing lists
  • Mesos/Splunk update:
    • Chetan/Tyson: on Cosmos DB:
      • Chetan: shares screen, short presentation
      • PR 3562 under main OW
      • Apache license does not allow Azure SDK, working with that other team to remove license hurdles to allow us to merge this PR
      • using this PR in some of or tests under our platform (replace CouchDB with Cosmos) similar constructs, document-level storage
      • activations can now be expired auto. by DB itself
      • as of now this PR, when JSON deps. fixed, can be merged pending comments from reviewers
      • next will work on attachment PR, then Cosmos can be supported soon.
  • Kubernetes: 
    • https://github.com/apache/incubator-openwhisk-deploy-kube/pulls?utf8=%E2%9C%93&q=is%3Apr+is%3Amerged
    • Dave Grove shares update on Helm charts for Kube
      • Helm subdirectory has good README includes:
        • 1-time Helm install.
        • deploy OW is simply 3 commands…
      • only takes a few minutes
      • Now updating Travis testing; goal is next 2 weeks to get 2 pieces of add. functional things from main OW not yet in Helm (e.g., Container Factory)
      • once done Helm will be primary way to install Kube
    • Dave shares perf. data
    • Carlos: is PR open around API GW (ingress config)?
    • Dave: it is “in flight” API GW talks to rest of system in a way not optimcal for Kube
      • might involve re-config
    • Dave: That is one test we cannot get working without some API GW change; may need to work on after rest of Helm work done
    • Carlos: Is playing around with installing this in IBM Cloud Kube service to make sure it works there
  • API Gateway
  • Catalog/Packages/Samples (anyone)
    • No update
  • Tooling/Utilities (Carlos/James/Matt/Priti, etc.)
    • https://github.com/apache/incubator-openwhisk-devtools/pulls
    • Priti: wskdeploy: “How do we specify credentials”?
      • new PR/solution is being reviewed/merged today
    • Priti: shows demo (using tests integrated into wskdeploy) to show how to declare inputs in the manifest at Project and Package level
    • Then shows how to interpolate values from ENV. VARs. and using Command Line parameter passing (individual and with file)
    • Basic overwrite of inputs similar to how CLI does it
    • New behavior (default) is if Env. Vars. are not filled in where expected in the Manifest, an error is shown.
    • Priti demonstrates new “Report” command, shows JSON preview of what will be sent to server (using OW APIs)
    • Matt will review today
    • Carlos: may need to look at scoping affect of param override from CLI
    • Carlos: good change for hiding parameter values from manifest


Confirm moderator for next call

  • Tyson will volunteer, for June 6th meeting
  • adjourn 11:01 AM US Central
  • No labels