Towards zero-downtime upgrades of stateful systems

Links
#1

https://stevana.github.io/towards_zero-downtime_upgrades_of_stateful_systems.html

4 Likes
#2

From the footnotes:

…I’m curious to know what your opinion is of:

#3

I haven’t read that paper, but I just skimmed it now and I agree with the
sources of complexity that they identify: implicit state, control and volume.

Their “Functional Relational Programming” is also similar to what I’m doing:

  • Their feeders and observers are my sources and sinks;
  • Implicit parallelism (pipelining);
  • They propose a relational model for transforming data that flows through this
    pipeline, while I haven’t thought of it that way it makes sense: one of my
    sources of inspiration is Jim Gray and in the video that I link to he said that
    this pipelining architecture is used inside relational databases.

I would have to study the paper in more detail to say more than that, unless you
have a more specific question? Either way, thanks for sharing the paper!

#4

My mistake: I should have been more specific - what is your opinion of section 5.2 (pages 15-18 of 66)?

#5

For what it’s worth, I also looked into module reloading at runtime for GitHub - ButlerOS/module-loader , but that didn’t work in non interpreted mode.

#6

There’s also GitHub - fbsamples/ghc-hotswap: Example code for how we swap compiled code within a running Haskell process. and GitHub - nmattia/haskell-hot-swap: Hot swapping compiled code while keeping a websocket connection open , I haven’t tried either yet though.

#7

This section seems to be mostly about the role of state in functional programming. I agree with the analysis there, it seems fairly standard and uncontroversial.