stream calculation output weird result

Nobody has claimed this yet.

I have a tickscript to calculate success rate based on the live data.
Here is my function:
success_rate = count('status'<400)/count('all');
The stream result ususally shows that the success_rate < 100%. But when I look into the origin data, all the data's status is 200. So that the success rate should be 100%.

Here is my script :
`var span = 1m`
`var frequency = 1m`
`var db = 'myna_publish'`
`var retentionPolicy = 'real_time_capture'`
`var sourceMeasurement = 'myna_publish_log'`
`var resultMeasurement = 'myna_publish_success_rate_per_1m'`

`var success_requests = stream `
` |from()`
` .database(db)`
`.retentionPolicy(retentionPolicy)`
`.measurement(sourceMeasurement)`
`.where(lambda: int("http_status") < 400)`
` |window()`
` .period(span)`
` .every(frequency)`
` |count('response_time')`
` .as('success_count')`

`var total_requests = stream`
` |from()`
` .database(db)`
` .retentionPolicy(retentionPolicy)`
` .measurement(sourceMeasurement)`
` |window()`
` .period(span)`
` .every(frequency)`
` |count('response_time')`
` .as('total_count')`

`success_requests`
` |join(total_requests)`
` .as('success', 'total')`
` .tolerance(1s)`
` .fill(0.0)`
` .streamName('success_rate')`
` |eval(lambda: if(float("success.success_count") / float("total.total_count") >= 1.0, 1.0, float("success.success_count") / float("total.total_count")))`
` .as('success_rate')`
` |influxDBOut()`
` .database(db)`
` .retentionPolicy('cq_retention')`
` .measurement(resultMeasurement)`

OS: Linux version 2.6.32-642.15.1.el6.x86_64 (mockbuild@c1bm.rdu2.centos.org) (gcc version 4.4.7 20120313 (Red Hat 4.4.7-17) (GCC) ) #1 SMP Fri Feb 24 14:31:22 UTC 2017
KapacitorVersion: Kapacitor OSS 1.5.1 (git: HEAD 89828ffff6cf5cd4cb2b34bf883e134395f734de)

Start by reproducing the supplied Tickscript with input whose http_status values are all 200, focusing on the window, count, join, fill, and eval stages. Trace the intermediate success_count and total_count values; the work is done when the cause of the lower success_rate is identified and the all-success case consistently reports 100%.

Written by the indexing model from the issue text.