2025 Q1 Roadmap

[Mr0grog]

We’re already halfway through Q1, but I wanted to write out for myself a quick roadmap of priorities and critical issues for the short term, since this project was in hibernation for a while and is now back in active use.

I’ve also made a GH project tracking all ongoing work. This document outlines critical priorities and buckets of work, while the project tracks detailed work in progress: Web Monitoring (view)

Overall Status

EDGI’s web monitoring efforts and most of the technical codebases here have been more-or-less shut down for the past 2+ years, but are now back in active use. The last month and a half has been a bit of a scramble to get everything running well again and working within our current constraints. Additionally, there is no technical team — it’s just me (@Mr0grog). We previously operated with several active technical team members, which influenced a lot about how these codebases are broken up and organized. I don’t currently expect this to change.

Given that, prioritization is pretty important. There’s a lot that probably won’t get done or at least won’t get much focus.

Core Near-Term Goals

Critical Issues:

• Work out data sourcing with IA. We used to rely on them to do all the canonical crawling/archiving, but there have been a whole host of issues with how that’s been managed this time around. We are currently doing our own crawling in addition.
  • Support importing data directly from WARCs (our own crawls and as a more expedient way of loading IA data). (Warcs have useful content too web-monitoring-processing#858)
  • Automate and operationalize our crawls if needed. Need to script and schedule all the pieces I am doing manually M/W/F nights:
    • Keep seed lists updated.
    • Crawl seeds with Browsertrix-Crawler.
    • Resume crawls that died in the middle.
    • Upload crawl output to S3.
    • Upload crawl output to IA.
    • Import crawl data to web-monitoring-db.
    • Analyze logs for errors that need to be handled specially because the server is not responding over HTTP (see below).
• Automate generation of weekly task sheets. This needs to not rely on me being around and able to run the script.
  • Add ability to auto-upload results to GDrive for analysts. (Auto-upload results to Google Drive web-monitoring-task-sheets#19)
  • Schedule automated runs 1/week. This is a little complicated because it needs a beefy box, and can’t just run on Kubernetes cluster. Two possibilities:
    1. Dockerize and run with AWS ECS, scheduled via AWS EventBridge alarms. (Dockerize this so it can run as a scheduled AWS ECS web-monitoring-task-sheets#6 + some ops work — I’ve done this kind of setup before with Terraform Cloud, but this will be more manual.)
    2. Run it as a GitHub action. My main concerns here are: a) requires a beefy box; is a GH runner up to the task? b) we save some costs by running inside AWS in the same region as our S3 bucket with archived response bodies; running in Azure (where GH actions are) may incur some new costs. But this is a lot simpler! (Add GH workflow for running analysis web-monitoring-task-sheets#18)
• Re-evaluate current deployment for cost optimization. Can we arrange things differently in AWS? Can we use reservations to get cheaper pricing on some resources? etc. Funding is overall down, and this infra is not free.
• Record versions where the server is no longer responding via HTTP (could be DNS resolution issues, TLS issues, server prematurely closing connections, etc.). We previously had some loose ways of recording page load errors when we were based on Versionista and PageFreezer, and losing the ability to discover and record network errors was something we hadn’t realized we were loosing when we switched to pure Internet Archive as a data source (we had stopped seeing servers just disappear at that point). Now that we’re seeing servers disappear again, this is a big issue.
  • Support recording these network errors in the DB/API. (Add network_error field to Version model web-monitoring-db#1184)
  • Write a tool to check for these (since we don’t proactively find out about them from IA) and import them to the DB.
    • I am currently adding records by hand with the source type edgi_statuscheck_v0 after examining the logs from our crawls.
    • A tool will need to be slow and careful about how it checks (maybe make repeated checks) since these errors are as likely to be bot blocking, firewall rules, proxy issues, etc. as they are to be legit servers going down.
    • The tool could check the crawl logs for candidates like I am currently doing (as long as we keep doing our own crawls).
    • Or it could schedule checks based on seeds with no data coming from IA when we do nightly imports.
    • Or it could check the DB for active pages with no new versions in the last N days.
    • Or some separate system for listing things it should check?
  • Schedule the tool to run automatically.

Cleanup:

• Remove all the maintenance banners on the various project repos, since work is active again.
  • Drop maintenance banner #176
  • DB: Drop maintenance banner web-monitoring-db#1203
  • UI: Drop maintenance banner web-monitoring-ui#1065
  • Processing: Drop maintenance banner web-monitoring-processing#862
• Clean up docs a bit; try and make them more current.

Important but Non-Critical:

• Alert analysts of possible new pages (and maybe removed pages?). This has always been a significant need that’s never been solved well. Some ideas we could do independently or together:
  • Keep track of seen URLs from all the links in all the pages we monitor. (Track seen URLs at all the domains we monitor #173)
  • Keep track of sitemap.xml for sites that have them (e.g. epa.gov). It would be neat to turn these into a git repo so they are also browsable through time, but that’s not the core need.
• Bring web-monitoring-diff back up to date and off life-support.
  • Wrap up the small refactor I started two years ago before everything shut down. It’s kind of in the way of other work. (Light cleanup on html_render_diff.py web-monitoring-diff#145)
  • Replace cChardet, which is blocking us from compatibility with current versions of Python. (Replace cChardet with something compatible with current Python versions web-monitoring-diff#165)
  • And/or support installations without cChardet. (Make cChardet and html5-parser (and lxml?) optional web-monitoring-diff#196)

Plus ongoing bug-fixing for analysts as they find problems.

Other Nice Stuff or Ideas

Not sure it’s likely I’ll get to any of this given how big the core stuff above is. But some things on my mind:

• Minimize web-monitoring-ui or even merge it into web-monitoring-db.

  The way these are split up makes a lot of things very hard to do or work on, makes infrastructure harder (more things to deploy, more CPU + memory requirements), and invites all kinds of weird little problems (CORS, cross-origin caching, login complexity, etc.). We originally designed things as microservices because of the way the team was structured and the skills people had, but that turned out to be over-ambitious in practice (in my opinion). Today, it’s an even more active problem when just one person is maintaining them all.

  We also had all kinds of ambitious ideas about the UI project giving analysts a direct interface to their task lists, being able to post their annotations/analysis directly to the DB, and so on. This never got done, and would require a lot more work, both in the UI, and in ways for the analysts to get their data back out or query it, before it would ever be better than the analysts working directly with spreadsheets as they do today. As things stand today, this stuff would be neat, but I don’t it is ever going to get done.

  If we drop all these ideas, the UI really doesn’t need to be nearly as complex or special as it is. It also doesn’t need its own server. At the simplest, it could be served as a static site (via GH pages, from CloudFront/S3, or even just from the API server as an asset). It could also just be normal front-end code in the web-monitoring-db server, but that requires a lot more rewriting and rethinking (it does pave a nicer, more monolithic path back towards including annotations/analysis forms for analysts, though).

  This would be some nice cleanup, but could turn into a big project. So a bit questionable.

• Consider whether web-monitoring-db should be rewritten in Python, and be more monolithic. The above stuff about merging away web-monitoring-ui feeds directly into this. Web-monitoring-db is really the odd duck here, written in Ruby and Rails while everything else is Python (or JS if it’s front-end). This was originally done because the first stuff I helped out with at EDGI was Ruby-based, and I thought there was crew of Ruby folks who would be contributing. That turned out not to be true. I think Rails is fantastic, but the plethora of languages and frameworks here has historically made contributing to this project very hard. Rewriting it in Python also makes it easier to pull other pieces (e.g. the differ, all the import and processing scripts, all the task sheet stuff) together, and would reduce some code duplication.

  I don’t expect this to go anywhere — this project is probably much too big and unrealistic at this point. But I want to log it.

• Get rid of Kubernetes. It’s been clear to me for several years now that managing your own Kubernetes cluster is not worthwhile for a project of this size. (I’m not sure it’s worthwhile for any org that cannot afford a dedicated (dev)ops/SRE person to own it.) Managed Kubernetes (AWS EKS, Google GKE, etc.) is better, but also still tends to be more complicated and obtuse than an infrastructure provider’s own stuff (e.g. AWS ECS+Fargate).

  This is also a big project on its own that probably won’t happen. Additionally, it’s possible it could be more expensive than the current situation (we have our services very efficiently and tightly packed into 3 EC2 instances, and you can’t make decisions that are quite as granular on ECS, for example), although there are other management tradeoffs.

  Note that a simplified, more monolithic structure as discussed above also makes it easier to run this project on other systems/services/infrastructure types. BUT we are probably somewhat coupled to AWS at this point, where all our data is.