different user cost different time by the same sql

I use InfluxDB 1.8.3, find a very strange bug.
in the same machine, i use SHOW TAG KEYS FROM measurement to query the tag keys.
but there is a huge gap when i use different user to query it.

1. use boss_metrics_read user, the user only have read privilege.

curl -s -w "time_connect: %{time_connect}\ntime_starttransfer: %{time_starttransfer}\ntime_total: %{time_total}\n" -GET 'http://localhost:8086/query?pretty=true&u=boss_metrics_read&p=xxxxxx' --data-urlencode "db=service_metrics" --data-urlencode "epoch=ms" --data-urlencode 'q=SHOW TAG KEYS FROM "service_server_rpc_v2_metrics"'
{
"results": [
{
"statement_id": 0,
"series": [
{
"name": "service_server_rpc_v2_metrics",
"columns": [
"tagKey"
],
"values": [
[
"class_method_name"
],
[
"class_name"
],
[
"client_instance"
],
[
"client_service"
],
[
"pkg_name"
],
[
"server_instance"
],
[
"server_service"
],
[
"traffic_type"
]
]
}
]
}
]
}
time_connect: 0.004
time_starttransfer: 12.113
time_total: 12.114

2. use boss_metrics user, the user have the admin privilege. which is the daily use user when i write the data.

curl -s -w "time_connect: %{time_connect}\ntime_starttransfer: %{time_starttransfer}\ntime_total: %{time_total}\n" -GET 'http://localhost:8086/query?pretty=true&u=boss_metrics&p=xxxxx' --data-urlencode "db=service_metrics" --data-urlencode "epoch=ms" --data-urlencode 'q=SHOW TAG KEYS FROM "service_server_rpc_v2_metrics"'
{
"results": [
{
"statement_id": 0,
"series": [
{
"name": "service_server_rpc_v2_metrics",
"columns": [
"tagKey"
],
"values": [
[
"class_method_name"
],
[
"class_name"
],
[
"client_instance"
],
[
"client_service"
],
[
"pkg_name"
],
[
"server_instance"
],
[
"server_service"
],
[
"traffic_type"
]
]
}
]
}
]
}
time_connect: 0.004
time_starttransfer: 0.009
time_total: 0.009

Is that the tag keys is not cached in the memory? Why the query time is so incredible，hope to reply soon!
Thank You!

The report targets the /query endpoint and SHOW TAG KEYS in InfluxDB 1.8.3; start by tracing how this query is handled for read-only versus admin users and reproduce both curl requests. Done means the cause of the timing difference is identified and the behavior is corrected or documented with a regression check.