Why is the probability of pod increasing after IPtable

Project clusters use k8s Managed , Traffic distribution uses iptable, According to the pressure measurement feedback, the load is unbalanced , I was so scared that I hurried to check , It turns out that the pressure test student made a mistake . During the query , Find out iptable After rule probability More and more big , Record the reason .

[email protected]:/home/admin# iptables -t nat -nL
...
Chain KUBE-SVC-7XZINH2IMK6FHKPK (2 references)
target     prot opt source               destination
KUBE-SEP-TOX5PBICM2IS3QEP  all  --  0.0.0.0/0            0.0.0.0/0            statistic mode random probability 0.20000000019
KUBE-SEP-BHWNMZ5XNYG7AESG  all  --  0.0.0.0/0            0.0.0.0/0            statistic mode random probability 0.25000000000
KUBE-SEP-OQLKQS2OMOJTYZF6  all  --  0.0.0.0/0            0.0.0.0/0            statistic mode random probability 0.33332999982
KUBE-SEP-EQOKMESPFMECIZJJ  all  --  0.0.0.0/0            0.0.0.0/0            statistic mode random probability 0.50000000000
KUBE-SEP-GVLV4E5D2AGR4O5J  all  --  0.0.0.0/0            0.0.0.0/0
...

You can see , in total 5 Bar record , At first probability by 0.2—>0.25—>0.33—>0.5— nothing , Why is that? .
Because after the flow enters , according to iptable Rule forwarding , public 5 strip , So the first reception 20% Of traffic .
There are still left. 4 strip , The next one should receive the remaining traffic 1/4, That is to say 25%.
There are still left. 3 strip , The next one should receive the remaining traffic 1/3, That is to say 33%.
…
The last one, because it's the only one left , All traffic will go here , So you don't have to write probability.