Deployments

One account can hold one of these per person. The list shows yours.

Deployments                              [ ] Show everyone's

i-0ca1eb44c4d712d8a                                      IDLE 5h
running · m5.xlarge · x86_64

UP      COST    RATE        LOAD    MEM FREE   DISK   AT RISK   SESSIONS
5.8h    $1.11   $0.192/h    0 / 4   96%        6%     0         0

· it has been idle for 5h: CPU averaging 0.3%, no sessions, nobody attached

3 repositories, all in GitHub. Safe to stop.

A list that mixes yours with a colleague's is a list where somebody stops the wrong machine, so yours is the default. Show everyone's is there when you need it, and labels whose each one is.

Machines nobody has claimed always appear, in both views. A filter that quietly drops one is how somebody concludes theirs is gone and launches a second.

What it costs

Computed, not billed. Rate times running time, using the on-demand price for that instance type in that region.

Cost Explorer has day granularity and hours of lag, which is no use for a machine that has been up for three of them. This is the only answer available in the moment, and it excludes the volume, the snapshots and data transfer.

A stopped machine still bills for its disk. That is not in this number.

Whether it is the right size

The verdict comes with its reasons attached, so you can disagree with it.

Pegged. The work is real and the machine is too small for it. That is a resize, not a restart.

Undersized. It is out of memory or paging to swap. A box at 40% CPU can be unusable because it is spending its time moving memory instead of working, and CPU alone would call that healthy.

Idle. Up, costing money, with nobody attached and nothing running. This is the fault nobody notices: nothing is broken, and the bill arrives a month later.

Idle needs all three to be true. Low CPU on its own is wrong, because a session sitting at a prompt part-way through a task is idle by CPU and must not be called abandoned.

Stopping one

Stop runs the same check as Before you stop, even when you are looking at somebody else's machine. Especially then: their unpushed work is the work you cannot see.

It stops. It never terminates. Stopping keeps the volume and everything on it.

How you know it failed

"rate unknown" in the cost column. The pricing API did not answer for that instance type and region. Uptime is still right; the money is not, and it says so rather than showing a number it made up.

A machine you expected is missing. Check the region, and check Show everyone's. It may be running under a colleague's name.