Leo Vantwo

Whatever you decide, run the tank dry in the autumn or put stabiliser in it. I have not had a spring no start since I started doing that and it costs about the price of a coffee a year.

Same database as the data, or an outbox. After-commit if you accept the crash window. Never inside the transaction with an external queue.

How do you set it per project? I only ever found the one setting in the account settings screen and assumed that was all there was.

Pretty sure invalid_client is about the client identifier, not expiry. An expired secret gives you invalid_grant. So if you are seeing invalid_client I would look at the Services ID first.

Learned this the hard way last month. Our wrapper reported "network error" for what was a perfectly well formed error response with a code in it. Two hours in the wrong direction.

Switched last year and the only thing I regret is losing the single bill. Not a technical complaint, just that I now have four small subscriptions instead of two and one of them renewed in a month I had forgotten about.

A lot of workplaces will let you leave it in a locker or behind reception if you ask once. Solved my problem for free after a year of assuming the answer would be no.

Right, and the same logic applies to your paywall and your feature flags. The first thing I regretted not making remotely configurable was the trial length, because I wanted to change it in week one and could not without a submission.

That is a good catch and worth checking first. Our generated code sits outside the module so it never showed up for us, but if yours is inside it would dominate everything.

Freeze the code Friday. Ship nothing the day before. The single biggest source of launch-week chaos is the confident small change made at 10pm the night before.

Three settings that measurably helped on our repo, in the order they helped. Turn off the extra static analysis passes, which run on top of the type checking and are the most expensive optional work gopls does. Turn off the code lens features you never click, particularly the run-test-above-every-function ones. And disable automatic imports scanning of the whole module if your editor is doing that on every keystroke.

After those three, peak memory on our repo dropped by roughly a third and completion latency became tolerable again. The analysis passes were by far the largest single contributor.

One thing to check when you pick the date: whether the US has flipped daylight saving relative to Europe. There are a couple of weeks each year where the gap is off by an hour from what you calculated, and it moves the whole schedule.

Which is also why the boring 'add a one line opt-out' advice matters more than people think. It's a cheap way to redirect the person who was otherwise going to hit report.

Sweep every minute for who is due. Unique index on user and date. Separate queue for bulk. Nothing has to happen at midnight.

One cheap thing while you decide: email the 63. Not a drip sequence, a single plain message a day after they stall, from your address, asking one question - what stopped you at the Google step. I sent 40 of those once and got 11 replies, and three of them said something I would never have guessed, which was that they were signed into three Google accounts and did not know which one the tool wanted. That was a one line fix in the UI.

Politely disagreeing with the get-the-floor advice. You do not need to talk much to be included, and chasing airtime is how people end up performing. What changed group settings for me was asking the person who just spoke one specific follow-up question. You are contributing, you are visible, you are not composing anything, and you get credit for being the person who is interested. I say maybe fifteen percent of the words in any group and get invited to everything.

Same fix here about a year ago. It also explains why it always seemed to happen after roughly the same interval.

Practical middle path we took: kept Node for the API surface, moved the one endpoint doing heavy work into a small Go service behind the same load balancer. Two weeks of work, no rewrite, and it let the team learn Go on something with a small blast radius. If the memory turns out to be concentrated in one code path, this is a much cheaper experiment than committing the whole service.

The simple-but-false thing is my whole experience of learning networking. Every clean analogy I was given had to be unlearned later, and the unlearning took longer than the original explanation would have.

Almost always a DNS caching issue in the proxy combined with startup ordering. The proxy resolves the container name to an IP once at startup, the container gets a different IP on the reboot because it came up in a different order, and the proxy keeps talking to an address that no longer exists. Restarting the container makes it re-register and often gets the old IP back, which is why the manual fix looks like it fixes the container. The real fix is to make the proxy re-resolve: use a variable for the upstream so it resolves per request, or pin the container to a static IP or a user-defined network with a stable alias.

Exactly the mechanism. Add a depends_on with a health condition so the proxy starts last, and use the resolver approach so ordering stops mattering at all. Belt and braces because reboots are the one time you are not watching.

That comparison had not occurred to me. Half the club is running splits off established hives.

Ten weeks sounds long until you compare it with the conversion rate of the alternative. Three customers from ten weeks of writing is a better hourly rate than most cold outreach I have run.

Table in Postgres, hosted service polling it, idempotency key on the row. Reach for a library when you have a third job type or a schedule.

The uncomfortable question: are they renewing because of the tool, or because of you? You said the tool is why they renew, but they are also paying $5,800 a month for someone who understands their reconciliation mess and answers the phone. Those clients are not evidence that a $99 product will sell, because they never bought a product, they bought your attention.

I am not saying do not do it. I am saying test the product on someone who has never met you before you restructure your income around it.