Skip to main content
This page is about the gateway itself, not the messages it carries. Nothing here counts messages, deliveries or vendor throughput — those are on the dashboard and Live health. This is whether the software has the memory, processing power and connection capacity to keep doing its job.
That separation is the whole design. A gateway can be delivering perfectly while quietly running out of file descriptors, and it can be idle while leaking memory. Neither shows up in a delivery figure.

​
Recent trend

Four sparklines — memory, processor, tasks running, work waiting — labelled since this page opened. The history is held in the browser tab and starts empty, so it genuinely cannot show you anything from before you arrived. For a longer view, point a monitoring system at the gateway’s own measurements endpoint.

​
The five verdicts

Each card carries a badge — Healthy, Worth watching, Under pressure, or Not reported — a summary, a fill bar, and two lines of prose: what the reading means, and the rule the badge was reached by. Publishing the rule is deliberate. A threshold you cannot see is a threshold you cannot argue with, and these are judgements about your platform, not laws of nature. If 70% processor is normal for your traffic shape, the card is wrong and you should know precisely how wrong.

​
What each one is actually telling you

Memory. Memory that keeps climbing and never falls back after a cleanup is how a leak looks. Run out entirely and the gateway stops accepting messages. The detail line counts cleanup pauses and the total time spent paused. Processor. Sustained high use means throughput is limited by this machine rather than by the vendor. Brief spikes during a burst are normal. Connection capacity. Every SMPP session, database connection and open file counts against this.
Reach the connection limit and the gateway refuses new vendor and customer connections before any other symptom shows. Nothing about the traffic looks wrong first. This is the card that fails quietly, and the reason it is on the page at all.
API requests. Messages submitted over HTTP, plus this panel’s own calls. Slow responses here point at the gateway or its database, not at a vendor. The figure is an average since the process started — useful for “this is slow in general”, useless for a spike a moment ago. Internal task queue. The gateway’s own background work — database writes, receipt handling. Work waiting here that never clears means it is arriving faster than it can be done, and the detail line will tell you if any was turned away outright, which is work the gateway could not do at all.
The internal task queue is not a worker pool. Worker means an SMPP connection everywhere else in this documentation and everywhere else in the panel. It is avoided here on purpose: readers who met the word on this card went looking for a vendor that was never involved.
Where a reading is simply absent, the card says so in words — “The gateway did not report a memory ceiling” — rather than showing a zero that reads like a measurement.

​
The raw readings

One control, Show the raw readings, opens everything the gateway publishes, in the software’s own names, grouped by subject. This is the same data a monitoring system would scrape, and it is here for investigating something the cards above cannot explain. There is a single search box across every group rather than one per card, because when you are searching you generally do not know which group the measurement is in — that is most of why you are searching. Groups matching nothing disappear rather than leaving empty headings behind. The step down is one step on purpose. An earlier version dropped a reader from plain English to raw label syntax in a single click, which is not progressive disclosure so much as a cliff. A footer reports when the page last read, and refreshes every ten seconds.