Help for owners
My site crashed during a sale even though I was told it scales automatically
Short answer
Automatic scaling adds servers only after a measured signal moves, and new servers need time to start. A sudden sale can outrun that, or hit a part that does not scale, such as the database. Ask your developer for the timeline of the crash and plan the next sale with a test, not a hope.
Does this sound like your situation?
- The site slowed or went down within minutes of an email or advert going out.
- My hosting dashboard shows more servers appearing, but only after the worst was over.
- I was told the site would scale automatically and it did not.
- New servers were added but the errors continued.
- Checkout failed while the home page still loaded.
Why does this happen?
Scaling reacts to a metric, and the metric lags
Dynamic scaling adds servers when a measured value, such as processor use, passes a threshold. The measurement and the new server both take time, so a spike that arrives in a minute can land before extra capacity does.
New servers need time before they help
AWS describes a default instance warmup: the time a new instance needs to finish initialising before its usage counts. It also says this is not enabled by default and that, without it, scaling policies can behave unpredictably compared with the real workload. Heavy start-up work, such as downloading and caching large files, makes this longer.
Confirmed by docs.aws.amazon.com
Only the web servers scale, not the rest
Extra web servers do not help if every one of them waits on the same database, payment service or file store. The sale then fails at the shared part.
The scaling limits were set too low
A maximum number of servers, or a cloud account limit, caps how far the system can grow, however high the traffic is.
Servers were added but failed their checks
If new servers are marked unhealthy by the load balancer, they never receive visitors.
Technical detail: Application Load Balancer returns 502 for a Node.js target
What can I do about it today?
- Write down the timelineNote when the campaign went out, when the first errors began and when it recovered. This is the most useful fact for finding the cause.
- Ask your developer five questionsWhat signal triggers scaling, how many servers can it add, how long does a new server take to be ready, what is the busiest shared part, and was it tested with load.
- Do not repeat the same promotion yetAnother large send into the same setup will most likely fail the same way. Stagger emails and adverts until the cause is known.
- Keep orders safeCheck that orders placed during the crash were recorded and paid correctly, and contact affected customers yourself.
- Do not switch off the scalingLeave the existing settings alone until someone has looked at them; changing them in a hurry can make the next spike worse.
When should I bring someone in?
- A big sale, launch or seasonal peak is coming and the last one failed.
- Nobody can explain what scaling was set up.
- The database or payment step failed even when the servers were added.
How do I handle it?
- Rebuild the timeline from the logsTraffic, server count, error rates and database load, side by side, show which part gave way first.
- Find the real bottleneckI test where the system fails under load before buying more servers, because more of the wrong thing does not help.
- Set up scaling to match start-up timeTrigger levels, minimum servers ahead of a planned sale, start-up work and warmup are set against measured values.
- Test before the next peakA load test at the expected volume, with a written rollback, comes before the sale, not after.
The price depends on what I find, so I quote in writing after a short discovery call. Tell me what is happening and I reply within 24 hours.
What do owners ask about this?
Can auto scaling make a site crash-proof?
No. It adds capacity in response to a signal, with a delay, and only for the part that is set to scale.
Should I just keep more servers running all the time?
For a planned sale that is often the simplest safeguard, but it costs money while it runs, so size it from a test rather than a guess.
Does AWS set up warmup by itself?
AWS states the default instance warmup is not enabled or configured by default, and recommends enabling it.
Sources
How was this page made?
I drafted this page with AI assistance from the official documents listed under Sources. Found a mistake? Tell me.