fix(c/driver/sqlite): Wrap bulk ingests in a single begin/commit txn

[ywc88]

Fixes #466 by having a single begin/commit txn for ingesting tables, instead of committing once per row.

Testing

R

I first installed the SQLite R driver and measured the time it takes to bulk ingest the nycflights13 dataset, noting the 336776 rowcount:

library(adbcdrivermanager)
db <- adbc_database_init(adbcsqlite::adbcsqlite(), uri = "test.db")
con <- adbc_connection_init(db)

flights <- nycflights13::flights
flights$time_hour <- NULL

stmt <- adbc_statement_init(con, adbc.ingest.target_table = "flights")
adbc_statement_bind(stmt, flights)

start_time <- Sys.time()
adbc_statement_execute_query(stmt)
end_time <- Sys.time()
[1] 336776

As well as the time it takes to execute the query:

end_time - start_time
Time difference of 1.711345 mins

As a followup, I noticed it takes significantly longer (~30 minutes) to execute the query on my XPS 15 9520:

• Ubuntu 22.04.2
• kernel 5.19.0
• i9-12900HK @ 4.90 GHz
• Intel Alder Lake-P
• 64 GB RAM

Compared to my Macbook Air (1.711345 minutes):

• macOS Ventura 13.4.1
• Apple M2
• 16 GB RAM

Both on the same version of R 4.3.1 (Beagle Scout).

After making my changes, I ran R CMD INSTALL . --preclean in arrow-adbc/r/adbcdrivermanager. I also installed the following R packages:

# install.packages("devtools")
# install.packages("pkgbuild")

After which I ran the following commands to validate no build / compile issues showing up as R packaged my changes:

devtools::build()
devtools::check()
devtools::install()

Noting that the file I made changes to showed up as a vendored file:

Vendoring files from arrow-adbc to src/:
- ../../adbc.h
- ../../c/driver/sqlite/sqlite.c
- ../../c/driver/sqlite/statement_reader.c
- ../../c/driver/sqlite/statement_reader.h
- ../../c/driver/sqlite/types.h
- ../../c/driver/common/utils.c
- ../../c/driver/common/utils.h
- ../../c/vendor/nanoarrow/nanoarrow.h
- ../../c/vendor/nanoarrow/nanoarrow.c
- ../../c/vendor/sqlite3/sqlite3.h
- ../../c/vendor/sqlite3/sqlite3.c
All files successfully copied to src/

After packaging and installing my changes, I ran through the same bulk ingest commands for the nycflights13 dataset and verified that the table contained the same number of rows as the previous run, also noting the speedup from 1.7 minutes to 0.2 seconds:

...
> start_time <- Sys.time()
adbc_statement_execute_query(stmt)
end_time <- Sys.time()
[1] 336776
> end_time - start_time
Time difference of 0.2236128 secs

[lidavidm]

LGTM

Is it possible to add a short test? Even in R or Python?