🎧 Kaizen 10 - The best pipeline ever™️ - Jun. 2, 2023

[gerhard]

🎧 This is now available as changelog.com/friends/2

See all improvements in this 🐙 GitHub repository: Kaizen 9 ... 10

---

This contains all things related to the next Kaizen episode.

Share your thoughts via comments below:

• Is there something in changelog.com that you, our dear listener, would like us to tackle?
• Is there something that you would like to help out with?
• Anything that we - @adamstac @jerodsanto - want to improve?

This is a rough timeline that we have in mind (red line is today):

gantt
    title 🎧 Kaizen 10 timeline
    dateFormat  YYYY-MM-DD
    axisFormat %b %e, %Y

    section DISCUSS
    ✍️ Add your comments below :2023-03-07, 77d

    section CONTRIBUTE
    💻 ALL CONTRIBUTIONS WELCOME 👕 :2023-03-07, 77d

    section RECORD
    🎙 Record :milestone, 2023-05-23
    ✂️ Edit :2023-05-23, 10d
    🚀 SHIP IT! :milestone, 2023-06-02

    section NEXT
    Kick-off Kaizen 11 :2023-06-02, 14d

Loading

---

🎙 Proposed topics

• Migrate Dagger v0.1 to Dagger Go SDK v0.5 #454
  • Removed the last Makefile
• Upgrade runtime deps - April, 2023 session #455 - Erlang, Elixir & Node.js + flyctl & Dagger Go SDK
  • 20 mins - including local CI testing
  • 🙏 Thank you @jamietanna!
  • BTW: .tools-versions via asdf-vm - Manage multiple runtime versions with a single CLI tool
• Read versions from .tool-versions when building production image #464
• Let me show you something cool
  • Changelog weekly issue #451 • 2023-04-09 - the last one
    • martinheinz.dev - Make Your CLI Demos a Breeze with Zero Stress and Zero Mistakes
    • saschagrunert/demo - A framework for pre-recorded CLI demos

brew upgrade && brew install dagger/homebrew-tap/dagger
dagger version
# must be >= 0.5.3
export _EXPERIMENTAL_DAGGER_TUI=inline
dagger run mage ci

• Typesense
• Fly.io Reliability: It’s Not Great - Reliability, it is great for changelog.com. @gerhard to pull stats from Honeycomb.

• Send app traces to Honeycomb #456

---

What changed so far? Kaizen.9 ... Kaizen.10

• Migrate Dagger v0.1 to Dagger Go SDK v0.5 #454
• Upgrade runtime deps - April, 2023 session #455
• Send app traces to Honeycomb #456
• W3c html validation fixes #462 🙏 Thank you @ken-kost
• Read versions from .tool-versions when building production image #464

What else is coming up?

• Add new caching variant for multi-node clustering #451
• 1Password Connect Server on Fly.io
• Honeycomb SLOs replace Pingdom monitoring
  • Uptime Kuma experiment
  • "Changelog Interviews" pods do not play at all on the website #457 (comment)

---

This is a follow-up to:

• 🎧 Kaizen 9 - Embracing change 🌟 - Mar. 2, 2023 #440
• 🎧 Kaizen 8 - 24 improvements & a lot more - Nov. 23, 2022 #433

[jerodsanto]

My main goal for this Kaizen is spinning off Changelog News into its own podcast + newsletter that replaces Changelog Weekly. This requires some technical changes/improvements to the app as well as a migration that will also require some single-use code. I feel confident that this will get done prior to Kaizen 10.

My secondary goal is to land on and execute a new plan for our feeds (I discussed this some on Kaizen 9 and the discussion with listeners of that is ongoing) that allows for:

1. Consistently fast enough responses
2. Clustering of app servers
3. Instant updates on episode publish/edit
4. Scaling from 10ish feeds to hundreds or low thousands

I feel not as confident that this will get done prior to Kiazen 10. In fact, I might just wait until we have Honeycomb visibility into app responses before I make serious moves on this effort...

In addition to all that, we have the version 0 of our Typesense integration finished. We should probably have Jason Bosco on (at least for a subset) to discuss his work on Kaizen 10.

[gerhard]

I am starting to look into how to send app logs as events to Honeycomb.

[gerhard]

WDYT? #456

[gerhard]

I know that we have replaced Algolia with Typesense for search.

Should we remove all Algolia-related code & secrets @jerodsanto? FWIW: https://github.com/search?q=repo%3Athechangelog%2Fchangelog.com%20algolia&type=code

Since we are talking about search, one thing that I would really like us to improve is the ordering of the search results. I would find most recent entries at the top most useful. For example, for https://changelog.com/search?q=llm, I would have expected for LLMs break the internet to have been the first search result, but instead this is what I'm seeing:

There are a few more search-related improvement suggestions that I have. Limiting myself to one to see how this goes. cc @jasonbosco

[jerodsanto]

We can certainly delete the Algolia code. Search improvements would be great too.

[jasonbosco]

@gerhard @jerodsanto Starting a new thread from here to talk about search relevance:

https://changelog.com/search?q=llm

We currently consider a match on the headline attribute of a news item as more relevant than a match on the story or fragment attributes:

changelog.com/lib/changelog_web/controllers/search_controller.ex

Line 10 in 60fd7eb

query_fields = "headline,story,fragment"

This is why a post with the title An LLM playground you can run on your laptop gets prioritized over the 2nd result which has LLM only in the fragment / story attribute.

If two news items have LLM in the headline and tie, then we use the published_at attribute to break the tie:

changelog.com/lib/changelog_web/controllers/search_controller.ex

Line 16 in 60fd7eb

sort_by: "_text_match:desc,published_at:desc",

So in your example, to get the 2nd result to rank higher based on recency, we would need to set all attributes to have the same weight...

[jerodsanto]

@jasonbosco sorry just getting back around to this. It appears the issue is that LLM doesn't match in the headline for LLMs break the internet, I'm assuming because of the trailing s?

If you run this search, it comes up first as it should (because it's both newer and has a headline match than the an LLM playground post).

https://changelog.com/search?q=llms

Do non-exact word matches like the above not factor in at all?

[jasonbosco]

It appears the issue is that LLM doesn't match in the headline for LLMs break the internet, I'm assuming because of the trailing s?

Do non-exact word matches like the above not factor in at all?

Non-exact (prefix) matches are indeed accounted for. For eg, if you search for LLM and scroll down by about 11 results, you'll see matches for LLMs show up, including the LLMs break the internet result. All of the results above that one have an exact match for LLM, which is why they are ranked first.

So the key thing is that exact matches are considered better matches and are ranked first, above non-exact partial matches.