`dbReadTableArrow` is much slower than `dbReadTable`

[avhz]

Using dbReadTableArrow seems to be much slower than the standard dbReadTable.

table <- DBI::Id(schema = "...", table = "...")

conn <- odbc::dbConnect(
    drv = odbc::odbc(),
    .connection_string = "..."
)

system.time(DBI::dbReadTableArrow(conn, table))
system.time(DBI::dbReadTable(conn, table))

> system.time(DBI::dbReadTableArrow(conn, table))
   user  system elapsed 
  1.623   0.020   1.757 

> system.time(DBI::dbReadTable(conn, table))
   user  system elapsed 
  0.706   0.005   0.727 

Note: this is from a SAP HANA database, connected via their odbc driver libodbcHDB.so.

[krlmlr]

Thanks. This is expected, until the odbc package implements this. The fallback is to use dbSendQuery() and to convert the resulting data frame chunks to Arrow.

You can still use this interface today to take advantage of free performance improvements once this is implemented, if you want to consume the data in Arrow format.

[avhz]

Ah thanks for the speedy response :)
Is there a timeline for this ?

Edit: I need the data in tibble format, so will cast from Arrow to tibble anyway, I would just like the fetches themselves to be faster. This would be the case once odbc implements it right ?