Skip to content

Performance ​

Settings → Performance shows how loaded the server is right now — live, without a look into the hosting platform. It answers the question "why is the app slow?" with one sentence and the figures behind it.

The page asks the API every 5 seconds while it is open (not while the browser tab is in the background) and shows the last 15 minutes. Nothing is stored: the figures live in the memory of the running API and start again with every deploy.

The verdict ​

The line at the top names the one bottleneck — or says that there is none:

It saysWhat it meansWhat to do
The container is being throttledThe API container may use only a fixed share of a CPU core (its CPU limit). Once it has used that share, the system pauses it until the next time slice — and while it is paused, every request waits.Raise the CPU limit of the API container in the hosting platform.
The CPU is at … of its limitThe container is close to its limit; more load and it is throttled.Consider a higher CPU limit.
The server is busyA request waits before the server even looks at it, although the container is not throttled: one long-running task is holding up the others.Look at the slowest requests below.
Memory is at … of its limitA container that reaches its memory limit is stopped and restarted.Raise the memory limit.
The database answers slowlyThe smallest possible question to the database takes long.Check the load and the limits of the database container.
No bottleneckThe container has headroom, the server answers at once, the database is fast.—

Why a low CPU limit hurts more than the number suggests

A limit of a quarter core does not make everything four times slower "on average" — it makes the server stand still for three quarters of every time slice as soon as there is work. And because the API answers all requests on one thread, a single expensive request delays everyone else's clicks too. That is why the Throttled tile can be critical while the CPU tile still looks harmless: the average use is low, but the work comes in bursts.

The tiles ​

Each tile shows the current value, the average of the last minute in its caption, its state in words (OK, Elevated, Critical) and a small trend line of the last 15 minutes.

TileShows
CPUThe CPU cores the container uses, against its limit (a bar) — or without a bar when no limit is set.
ThrottledThe share of the time the container was paused by its CPU limit. 0 % is the goal; from 5 % on it is noticeable, from 25 % on critical. Without a CPU limit a container is never throttled.
Server response delayHow long a request waits before the server looks at it (the delay of the server's event loop). A few milliseconds are normal.
MemoryThe memory the container uses, against its limit.
RequestsRequests per minute, how many are being answered right now, and how long requests took in the last 15 minutes (half of them less than …, 95 % less than …).
DatabaseHow long the database takes to answer, its size, the number of snapshot rows and the connections in use.

Slowest requests ​

The requests of the last 15 minutes that took the longest, grouped by kind (for example GET /weclapp/customer-billing): how often, the average, the time within which 95 % were answered, and the slowest one. Only the kind of a request is recorded — never an address with a customer's name in it. Lists that the app asks for in one bundled request (several small lists that a page needs at the same moment travel together) are counted here one by one, each by its own kind.

The line below the table names the API version, how long it has been running, and the machine under the container (CPU cores, load, free memory).