always-decrease-rw-together = true ignored

[doodoori2]

Hello
I'm using dynamic_dynamodb-2.2.1
always-decrease-rw-together = true ignored..
and it decrease read and write seperately.
consume decrease opportunity 2 times.

reads-lower-threshold: 40
decrease-reads-with: 25
decrease-reads-unit: percent
num-read-checks-before-scale-down = 10
num-read-checks-reset-percent = 80

writes-lower-threshold: 40
decrease-writes-with: 25
decrease-writes-unit: percent
num-write-checks-before-scale-down = 10
num-write-checks-reset-percent = 80

allow-scaling-down-reads-on-0-percent: false
allow-scaling-down-writes-on-0-percent: false
always-decrease-rw-together: true

more config (gist)

Changing provisioning to 3572 read units and 6500 write units
Updating provisioning to 3572 reads and 6500 writes

...

Consecutive read checks 9/10
Consecutive write checks 10/10
Changing provisioning to 3572 read units and 4875 write units
Updating provisioning to 3572 reads and 4875 writes

...

Consecutive read checks 10/10
Consecutive write checks 0/10
Changing provisioning to 2679 read units and 4875 write units
Updating provisioning to 2679 reads and 4875 writes

more log(gist)

[mnietz]

Same here.

[reidfaiv]

Quick look on code: https://github.com/sebdah/dynamic-dynamodb/blob/master/dynamic_dynamodb/core/table.py#L102

The following code seems to cause it:

def __calculate_always_decrease_rw_values( ... ):
...
    if read_units <= provisioned_reads and write_units <= provisioned_writes:
        return (read_units, write_units)

and it should be instead:

    if read_units < provisioned_reads and write_units < provisioned_writes:
        return (read_units, write_units)

It appears it was like this before, but was changed by following commit: b4b1bc1
Based on this commit comment, its intention was to fix the case when either one read or write was already on lower limit and couldn't be scaled down any further. Probably code should compare against min-provisioned-reads and min-provisioned-writes values instead.