Miro moved 70,000 tests from Jest to Vitest after years of slow CI, memory leaks, and failed patches

If Jest's 13 tests per second is a per-runner rate, the post's two CI figures agree with each other. Seventy thousand tests split across six runners is about 11,700 per runner ^[19]. At 13 a second that comes to roughly 900 seconds, or 15 minutes ^[20]. The author gives a ceiling of 20 minutes per runner ^[4]. On that reading, most of each job is Jest running tests, with SWC already doing the parsing ^[5].

The suite also got slower faster than it grew. In July 2024, six runners took around 7 minutes each for 35,000 tests ^[12]. The count then doubled to about 70,000 ^[17] while the per-runner ceiling rose about 2.9 times ^[21]. One figure is "around" and the other "up to", so the ratio is loose. I'd expect that shape from memory piling up in long-lived workers, and the post says a quarter of all runs fail because leaks kill runners with out-of-memory errors ^[6].

The earlier refusals were sound for their time. In November 2023 a colleague wrote that "the extra set of vitest specific configurations and replicating our current build might be too risky, without knowing how much speed benefits we get now that we're relying on @swc/jest" ^[9]. The team had just started running unit tests in parallel and was buying back wall time with extra runners, an approach that "Worked pretty well", Sergey wrote ^[10]. In July 2024 another colleague wrote: "I expect higher returns from better test selection (e.g. not running 35k unit tests each time). The cost of switching to vitest (for example) is high and unclear what the improvement is" ^[11].

Waiting for Jest to improve did not pay. The move from Jest 29 to Jest 30, the first major release in three years, gave about 5% on speed and memory ^[8]. Miro had already patched Jest core packages to get a better worker pool, and those patches were never accepted upstream ^[7].

Laptops broke as well. In March 2025 the VSCode Jest extension installed several versions of fb-watchman, Jest's filesystem watcher, and saturated engineers' CPUs until the team set `watchman: false` ^[13]. Ahmed, a principal engineer on the team, said: "I want to say maybe it's time to switch to vitest, but that's a massive effort..." ^[14]. Sergey had no capacity to lead it then, and with AI models getting better at complex problems, "All signs were to wait for a bit" ^[15].

The two-month duration is Sergey's claim, made in the post's headline ^[1]. The excerpt ends before the migration plan and the after-migration timings, so the duration is the only result on record. The starting state is specific. A dedicated Frontend Developer Experience team owns the tooling ^[2]. The monorepo runs on yarn and NX across more than 800 packages ^[16]. In the same year that group replaced Prettier with Oxfmt and ESLint with Oxlint, and moved CI typechecking to TypeScript 7 ^[18]. A team running 35,000 tests in seven-minute jobs is where Miro stood when a colleague argued for test selection over a switch ^{[11] [12]}.