Kai

Turn on trace: 'on-first-retry' and upload test-results/ as an artifact. Ten minutes of setup and it pays for itself the first time this happens while you are asleep.

If you want to reproduce it locally, run the suite pinned to two cores. Same starvation, on a machine where you can attach a debugger.

Do not forget the states between active and expired. Billing retry can run for weeks, and grace period keeps the user entitled while the payment is being retried.

If you flip access off the instant the expiry date passes, you cut off people whose card is being retried and who would have paid you. They then churn for real, because you told them the subscription was over. That is the version of this bug that costs actual money - three days of missing Pro is a support email, a wrongly revoked grace period is a cancellation.

Add one canary test that only passes when RLS is actually on. Otherwise the day someone runs CI as a superuser or a role with bypass privileges, every policy test goes green and stays green, and the suite is worse than useless because it is now lying confidently.

// @ts-ignore on that line and move on. TS2589 is the compiler being conservative, the type is fine at runtime.

Bag tip: prop it with a couple of chopsticks so the plastic is not touching the leaves, and open it for an hour a day. Otherwise you can end up with fungal spots on top of everything else.

Servers do not, deprecations do. A runtime going out of support, a TLS chain change, a payment API version sunset, a major database upgrade you cannot defer any more. "Forever" costs about a day a year even if literally zero customers email you, and it arrives on a schedule set by other people.

And include a demo account in the notes even when you are sure they do not need one. It costs you nothing and removes an entire category of second round.

Let the database decide, and put the side effect after the write wins.

const [row] = await db
  .insert(invoices)
  .values({ providerEventId: event.id, ... })
  .onConflictDoNothing({ target: invoices.providerEventId })
  .returning({ id: invoices.id })

if (!row) return   // someone already handled this event, ack and go home

await sendReceipt(row.id)

If no row comes back, this delivery is a duplicate and there is nothing to do. No exception, no retry, no second email.

The broader point: your retries are not the bug. At-least-once is the guarantee you bought when you picked a queue, and anything you put on it will run twice eventually - on redelivery, on a stalled worker, on a deploy. The handler has to be safe under that, always.

One thing that quietly slows everything down: nutrients. A brand new coco and sphagnum substrate is close to inert. Once the cleanup crew and some leaf litter have been breaking down for a few months, growth picks up noticeably by itself. Until then a very dilute foliar feed every few weeks helps, as long as you keep it weak enough to be safe for whatever will eventually live in there.

The power argument gets used to justify the Pi more than the numbers support. A Pi 5 idles around 4W, an N100 box around 7 to 10W. Call it 5W of difference, which at typical residential rates is somewhere near 6 to 10 a year. That is not the deciding factor for a machine you will keep for five years.

vi.mock calls are hoisted to the top of the module by the transform, above everything you wrote, regardless of where they appear in the file. That is deliberate - the mock has to be registered before the imports it replaces are evaluated. So the factory runs first and mockDb is still in its temporal dead zone.

Hoist the value with the mock:

const { mockDb } = vi.hoisted(() => ({
  mockDb: { query: vi.fn() },
}))

vi.mock('~/db', () => ({ db: mockDb }))

vi.hoisted gets lifted alongside the mock, so both the factory and your tests can see the same object. Rule of thumb: anything the factory touches must be created inside vi.hoisted.

Also check whether the new spot puts it near a radiator or in the draught from a door. Figs sulk about cold air on the leaves and the timing would look identical to a light problem.

Worth knowing before you tune anything: maxStalledCount defaults to 1, so a job that stalls twice is moved to failed rather than looping forever. That is why some of yours ended up failed instead of running six times - the framework already saved you from an infinite loop. Raise it only when you understand why they stalled.

The recirculate point catches a lot of people. It feels like it should keep the damp out, but it just recycles the wettest air in the car back over the glass. Fresh air plus AC clears fog several times faster.

This is also why you cycle for four to six weeks before the animal arrives. The bloom passes, the plants root in, the cleanup crew builds numbers, and you find out about any condensation or drainage problems while there is nothing living in there to worry about.

It is the Mpemba effect and physics still cannot explain it at all, it is one of those genuinely open mysteries.

If four to six weeks of manual treatment does not get you there, systemic granules watered into the soil will. They are genuinely effective on mealybugs because the insect drinks the plant. I would only reach for them after the manual approach has failed, and not at all if you have a cat that chews leaves or a toddler who puts things in their mouth.

Also check what you are measuring with and where. Cheap analog dial hygrometers are frequently off by 15 or 20 points and they respond slowly, so a reading of 40 might be a reading of 60 taken badly.

Use a digital probe, put the sensor at animal height rather than at the top of the tank, and if you can, calibrate it once with the salt test. Half the humidity panics I see turn out to be instrument problems.

It depends on one thing: does your data have a future in it.

A $19 notes app, 30 days is generous. A scheduling tool for a clinic has rows dated after your shutdown date - appointments people have already promised to patients. That is not "your service ends", that is "their next quarter needs rebuilding somewhere else, by hand, while they are working".

Practical version of the test: select max(starts_at) from appointments. Whatever that date is, your read-only date should be after it, or you should give people enough runway to move the future-dated rows. For most scheduling tools that lands somewhere between 60 and 90 days, and it is not about politeness.

Draining the tin was what fixed my chilli. I had been treating tinned tomatoes as one ingredient rather than as an ingredient plus most of a cup of water.

This is the part that gets skipped in every "which queue" thread. "Redis is faster" is true and irrelevant at 20 a minute. "Redis is not in your transaction" is the thing that costs you a weekend and a customer apology.

Two other things worth checking:

Leaf litter depth. Springtails need a damp layer of decaying material to live in, and if yours has broken down over two months and not been topped up, their habitat has quietly disappeared. I keep at least two inches of magnolia or oak leaves and add a handful monthly.

Which isopods you added. Powder oranges and dwarf whites are gentle and happy in the same damp conditions. Some of the larger species prefer drier setups, and keepers adjust the tank to suit them, which is what actually harms the springtails. If you changed your misting routine for the isopods, that is your culprit rather than the animals themselves.

Small thing that might help the pattern make sense: nothing about you changes much here. Exhaled air is at body temperature and saturated whether you are hydrated, out of breath, or freezing cold. So when the fog changes and you did not, it is the atmosphere doing it.

Whatever you choose, do the substrate barrier properly. Fine window screen or a layer of fibreglass mesh between the drainage layer and the substrate, cut to the exact footprint with no gaps at the edges. Substrate washing down into the drainage layer is the failure mode that eventually forces a teardown regardless of which method you picked.

The one size rule is the bit people skip and then wonder why the new pot stays wet for a fortnight. Two inches of extra diameter is plenty.