NEWS (original) (raw)

pool 1.0.4

Now explicitly requires DBI 1.2.0 (#178) and messages if you’re using an old dbplyr (#179).

No longer depends on the withr package, by instead requiring R 3.6.
Add wrappers for dbplyr generics db_col_types()(#171) and db_copy_to() (#172).
Pool no longer generates spurious messages about needing to usein_schema() or avoiding the use ofident_q().
Add support for new DBI generics that return Arrow objects.

copy_to() now returns a tbl that uses the Pool.
Added missing methods for sql_join_suffix() (#165) and sql_query_explain() (#167).

Pool has been re-licensed to MIT (#158).
dbPool() gains an onCreate parameter that allows you do something to every connection that pool creates. This is useful for setting options that you want to apply to every connection (#98).
New localCheckout() checkouts and then automatically returns an object. It only works in function scope.

Pools now get a useful print method (#140).
pool now implements the dbplyr 2.0.0 interface, eliminating warnings when using pool with dplyr (#132).
Pool errors and warnings have been reviewed with an eye to making them more immediately actionable (#145).
Objects are now validated once on first checkout to ensure that the object and validation strategy are both ok.
Added support for SAP HANA databases (@marcosci, #103).
dbPool() and poolCreate() now default to validating every 60s, rather than every 600s. This makes pools a little more robust to shorter connection timeouts (#149).
dbPool()’s validateQuery is now actually used (#153).
DBI methods should dispatch correctly in more cases; in particular dbReadTable() and friends will now work correctly when used with DBI::Id() (#120).

dplyr and dbplyr are now Suggests instead of Imports. Thanks,@AkhilGNair! (#106)
When used with dbplyr, tbls now store a copy of the pool, not a checked out connection. (#107)
dbListObjects(), dbCreateTable(),dbAppendTable(), dbIsReadOnly(),dbQuoteLiteral(), and dbUnquoteIdentifier()methods are now implemented for pool objects. (#100, #109)

Previously, pool would always set options(warn=1) when running tasks. It now ensures that the value of warn can be 1 or greater. This can be useful in debugging, so thatoptions(warn=2) can be used. (#90)

Changed the methods dbExistsTable(),dbRemoveTable(), dbWriteTable(),dbGetQuery(), dbExecute(),dbListFields() and dbReadTable() to dispatch over the first two arguments, as per the default definition in DBI. (#57)

Use requireNamespace("pkg", quietly = TRUE) forRMySQL and RSQLite in the examples and tests since these are “Suggests” packages (i.e. not “Depends”). (commit 4205feb)

Included more examples in the documentation. (#50)
Fixed the “test-create-destroy.R” test. Previously, this test had to be run manually because it uses later and its async nature could not be captured by testthat. However, usinglater::run_now() immediately after the relevant code snippet (i.e. still inside the first argument of expect_*) solves the issue. (#50)
Use difftime(t1, t0, units = "secs") when calculating a time interval. Unlike the simpler t1 - t0method, this guarantees that the result will always be consistently the number of seconds. However, there’s no change when calculating a new time (not a time interval) usingt2 <- t1 - interval, since we want t2 to be a time, rather than a time interval (which is always what is returned bydifftime). (#50 and #48, thank you @caewok!)

Fix all dbplyr wrapper functions that weren’t passing in additional arguments because the call to the originaldbplyr function included ... = ... instead of.... Also, pass temporary = temporary incopy_to.Pool, so that we don’t defeat the whole purpose of that wrapper. (#50)
Change the place where the check for the maximum number of objects is made. Previously, this chunk of code was misplaced and it would result in buggy behavior: namely, once the maximum number of objects was reached, no more objects could be checked out (even if you returned any/all objects back to the pool). The only reason this wasn’t spotted earlier is because the defaultmaxSize is Inf (and there’s usually not a good reason to change it). (#50)

Fix #39: Moved dplyr support in poolfrom dplyr 0.5.0 to dplyr 0.7.0, which includes a lot of breaking changes including the addition of a brand new package called dbplyr. (#42)
For pool users, the main change is that all thesrc_* functions are now gone (from dplyr andpool). Therefore, if you had something like:

data <- src_pool(pool) %>% tbl("test")

You can just change it to the simpler:

data <- pool %>% tbl("test")

If you’re still on an old version of dplyr and want to use pool as well, please install the package using the tag created for that purpose:

devtools::install_github("rstudio/pool@dplyr-pre-0.7.0-compat")

Changed all time arguments to accept number of_seconds_, instead of milliseconds. This is because this is what the later package uses and there was no reason for pool to be different, except backward compatibility. Since both time arguments to dbPool(idleTimeout and validationInterval) have default values, we’re hoping this change won’t even be noticed by most users. If you were setting either of those directly, however, you will need to update your app if you update the pool package. (#44)
Dropped the Pool methods around dbConnect anddbDisconnect, because these made it easier to lose track of whether you’re operating on a Pool object or on a database connection directly. From now on, only these allow you to get a connection from the pool and return it back, respectively: (#44)

con <- poolCheckout(pool)  
poolReturn(con)

Use later package for scheduling tasks (#44). This also has the side effect of fixing #40 and #43 since laterallows us to get rid of the naiveScheduler completely.