Source
CloudWatch AWS/ApplicationELB, .
ComfyUI is target group
Definitions
- Served by ComfyUI = ComfyUI RequestCount − ComfyUI HTTPCode_Target_4XX_Count. A 4XX
response was not completed by ComfyUI, so it is not counted as served.
- Served by faceswap = faceswap RequestCount. All requests = the two added, and each share
is of that total.
- Served by ComfyUI per hour = served by ComfyUI ÷ complete hours in the range.
- Instances = HealthyHostCount ÷ 8 (8 target ports per EC2 instance on both target groups).
The tile shows the mean of the hourly averages, with the minimum of the hourly minimums and
the maximum of the hourly maximums. The published report used the hourly averages for the
ComfyUI minimum; that one figure can read slightly higher there.
- Range latency = the unweighted mean of the daily TargetResponseTime datapoints for the
Philippine days that lie entirely inside the range and have already finished. A day the
range only partly covers is left out, and so is the current day. When no day qualifies,
the tile falls back to the mean of the hourly datapoints and says so.
Time
Everything is shown in Philippine time (UTC+8), and days are Philippine calendar days.
The daily latency datapoints are CloudWatch one-day buckets aligned to Philippine
midnight. A range starts inclusive and ends exclusive, so "Aug 20 – Sep 2" covers
Aug 20 00:00 through Sep 3 00:00.
Buckets in progress
A bucket whose end is still in the future is in progress. Totals always use complete hours.
In the daily view the current day is drawn dimmed with its complete hours so far and marked
"(in progress)" in the table.
Comparison
"vs previous period" compares the range against the period of the same length immediately
before it. Shares are compared in percentage points, counts and latency in percent, instances
in absolute instances.
Files
METHODOLOGY.md in this repository carries the full
method. The published PDF report is produced by the pipeline in
comfyui-api-handler/reports/traffic-shift from the same definitions.
