Batch insert in PostgreSQL extremely slow (F#) - postgresql

Question

The code is in F#, but it's generic enough that it'll make sense to anyone not familiar with the language.
I have the following schema:
CREATE TABLE IF NOT EXISTS trades_buffer (
instrument varchar NOT NULL,
ts timestamp without time zone NOT NULL,
price decimal NOT NULL,
volume decimal NOT NULL
);
CREATE INDEX IF NOT EXISTS idx_instrument ON trades_buffer(instrument);
CREATE INDEX IF NOT EXISTS idx_ts ON trades_buffer(ts);
My batches are made of 500 to 3000 records at once. To get an idea of the performance, I'm developing on a 2019 MBP (i7 CPU), running PostgreSQL in a Docker container.
Currently it will take between 20 and 80 seconds to insert a batch; while the size / time is not really linear, it scales somewhat.
I'm using this lib https://github.com/Zaid-Ajaj/Npgsql.FSharp as a wrapper around Npgsql.
This is my insertion code:
let insertTradesAsync (trades: TradeData list) : Async<Result<int, Exception>> =
async {
try
let data =
trades
|> List.map (fun t ->
[
"#instrument", Sql.text t.Instrument.Ticker
"#ts", Sql.timestamp t.Timestamp.DateTime
"#price", Sql.decimal t.Price
"#volume", Sql.decimal (t.Quantity * t.Price)
]
)
let! result =
connectionString
|> Sql.connect
|> Sql.executeTransactionAsync [ "INSERT INTO buffer_trades (instrument, ts, price, volume) VALUES (#instrument, #ts, #price, #volume)", data]
|> Async.AwaitTask
return Ok (List.sum result)
with ex ->
return Error ex
}
I checked that the connection step is extremely fast (<1ms).
pgAdmin seems to show that the PostGreSQL is mostly idle.
I ran profiling on the code and none of this code seem to take any time.
It's as if the time spent was in the driver, between my code and the database itself.
Since I'm a newbie with PostGreSQL, I could also be doing something horribly wrong :D
Edit:
I have tried a few things:
use the TimeScale plugin, made for time series
move the data from a docker volume to a local folder
run the code on a ridiculously large PostgreSQL AWS instance
and the results are the same.
What I know now:
no high CPU usage
no high ram usage
no hotspot in the profile on my code
pgAdmin shows the db is mostly idle
having an index, or not, has no impact
local or remote database gives the same results
So the issue is either:
how I interact with the DB
in the driver I'm using
Update 2:
The non async version of the connector performs significantly better.