Monitoring
The application dashboard — every running instance in one line of state, throughput and failures over a moving window, host and process resources, and what each instance occupies.
Monitoring is the dashboard for the whole application: every running instance added up, and each of them apart. It polls every five seconds and holds a moving window of 2.5, 10 or 30 minutes, chosen in the page header. With nothing running, the sections that depend on generation are replaced by a note pointing at Instances, while the host and process resources stay live.
The line of state
The row under the title is the application in one line: the instances running, the rate each stage of generation is moving events at, the rate at which events are failing, and how full the fullest queue of any instance is. The totals since those instances started close the row — timestamps generated, events produced, events written and events dropped.
| Reading | Meaning |
|---|---|
| Input | Timestamps entering the pipeline per second |
| Event | Events produced from them per second |
| Output | Events written to their destinations per second |
| Failing | Events lost per second, at the event stage and the output stage together |
| Fullest queue | The most loaded queue of any running instance, as a share of what it may hold |
Read left to right, the rates locate where events are going missing without opening a single instance: an event rate under the input rate is the event stage dropping or failing to render, and an output rate under the event rate is a destination refusing them.
Throughput and failures
The Throughput chart plots input against output as rates measured between polls — the rate now, rather than the average since a generator started that the Instances table reports.
The Failures chart covers the same window beside it, split into the event stage and the output stage.
Resources
CPU and memory for the host, disk and network for the Eventum process, each with its history over the window. The memory reading also names how much of the host's memory the Eventum process itself holds.
Read next to the throughput chart, these separate a slow generator from a saturated machine: output that plateaus against a CPU ceiling is a different problem from output that plateaus with the machine idle.
Instances
Everything about the running instances is one section: a chart of the load over the window, and the table of figures behind it.
By instance stacks the output rate of each instance, with the smallest of them folded into a single band, so an instance that dominates the pipeline — or has quietly stopped contributing to it — is visible in the shape of the chart. By stage stacks the same window by pipeline stage instead.
The search and the quick filters — Failing, At the limit, Idle — narrow the chart and the table together, and colour follows a row from one to the other. Selecting a row opens what that instance is doing beside it: the totals behind its rates, what each of its plugins moved, and the state of both its queues, one link away from its own page.
The table ranks the instances by what they occupy, and sorts by every figure it shows.
| Column | Contents |
|---|---|
| CPU | Share of one processor core over the last poll |
| Wait | Share of that time spent ready to run while waiting for a processor |
| Output | Events written per second |
| Failures | Events lost per second |
| Disk write | Bytes per second written through the file system |
| Network out | Bytes per second sent |
| Events queue | Memory the batches waiting between the event and the output stage occupy, against the limit they may |
| Threads | Threads the instance runs |
Wait is what separates an instance the machine cannot keep up with from a slow one: a figure that climbs across every instance means more of them are running than the host has processors to run, while a single hot instance next to calm ones names the generator to stop.
Outside Linux the operating system does not account for waiting or file system bytes per thread, so Wait and Disk write read as zero there. The rest are reported on every platform.