- switch from "unknown" parameter in text format to a "json" parameter in
binary format (no dependency update required)
- use a lazy bytestring "json" encoder (via updated hasql)
- hasql-pool-0.8
* 'release' now flushes the pool, and no longer destroys it;
'PoolIsReleasedUsageError' is gone compared to our fork
* now supports an acquisition timeout, which we don't use
yet
* lower bound on 0.8.0.2 to fix a Windows build issue
- hasql-1.6
introduces a position parameter to ServerError, which we
ignore
This version of hasql-pool is a simplified rewrite that doesn't use
the resource-pool package. The major API changes are that idle
connections are no longer timed out (and the corresponding setting
is gone), and that `release` makes the pool unusable, where it used
to remain usable and only flushed idle connections.
We depend on a PostgREST fork of 0.7.2 that gives us reliable
flushing, compare https://github.com/PostgREST/hasql-pool/pull/1
- hasql-pool 0.7 removes timing out of idle connections, so
this change removes the db-pool-timeout option.
Given that we were typically running with very high
timeout settings, I don't anticipate the lack of timeout
to introduce new issues, though we might want to consider
introducing some retry-logic down the line when we
encounter connection failures.
- See https://github.com/PostgREST/postgrest/issues/2422 for a
discussion on depending on a forked dependency. Besides adding
the dependency to the nix overlay, we're also adding it to
stack.yaml and a new cabal.project to allow stack/cabal users
to build the project.
Change callProc/createWriteStatement to H.Snippet.
Parametrize the inputs on the same SQLFragments by taking advantage
of hasql-dynamic-statements. It's not necessary to parametrize
every input, inlining with pgFmtLit can still be used for queries
that can't be parametrized(like SET LOCALs).
Also adds hasql-dynamic-statements to Nix and stack.
Fixes https://github.com/PostgREST/postgrest/issues/1512
Helps on environments where you can't send unix signals(Windows, managed
containers). Also provides better UX for schema reloads - NOTIFY
can be sent from pg clients(psql, pgadmin).
`NOTIFY pgrst` - with no payload - should be done to reload the schema cache.
Notifications with a payload will be ignored.
The channel can be enabled with `db-channel-enabled`(false by default)
and its name can be configured with `db-channel`.
The LISTEN thread uses a dedicated pg connection.
This connection is recovered if it fails.
A debounce of 1ms is done in case too many NOTIFYs arrive.
When doing:
```
cabal upload dist-newstyle/sdist/postgrest-7.0.1.tar.gz
```
The following error is shown:
```
Error: Invalid package
'ghc-options: -Wall -Werror' makes the package very easy to break with
future GHC versions because new GHC versions often add new warnings.
Use just 'ghc-options: -Wall' instead. Alternatively, if you want to
use this, make it conditional based on a Cabal configuration flag
(with 'manual: True' and 'default: False') and enable that flag
during development.
```
Put -Werror in a cabal flag to work around this restriction.
The good part is that protolude 0.3.0 builds with GHC 8.10.
The bad part is that this change is a bit painful:
- the default `toS` has changed to no longer convert to and from ByteString
- similarly, `show` no longer outputs ByteString
The changes here are pretty much minimal to keep things compiling; I didn't
see a nice way to work with the new ConvertText class, even though `toUtf8`
seems like it might help if used besides `toS` at just the right spots.
The changes here are from fixing stack build with LTS 15.8 (GHC 8.8.3)
to be warning-free. Most if not all come from the newer protolude
version, which adds some things like `unwords`.
So to make sure that we don't get unused import warnings with newer
stackage, this both raises the protolude lower bound and removes
these imports.
* Don't indent first line of configuration parser error messages
configurator-pg returns multi-line errors now. Alternatively,
we could indent all the lines.
Two changes:
- tags are now an insert-ordered hashset
- type_ .~ -> type_ ?~ with some magic type inference
This also updates stack.yaml to the newest GHC 8.6 LTS release.
* Update resolver to lts-13.29 and add lock file to repository
* Upgrade stack version
* Save cache after building dependencies only to have faster feedback loop when tests fail
* Move private functions from QueryBuilder to a separate Private module
* Move more functions over to private trying to make compilation consume less memory
* Split private in 4 modules
* Remove unused LambdaCase pragma
* Add profile to memory-tests.sh so it can find postgrest executable
* Move save dependencies before building and running tests for faster feedback loop
See also https://github.com/haskell/cabal/issues/4739.
/Users/travis/build/PostgREST/postgrest/<built-in>:15:10: error:
error: non-portable path to file '".stack-work/dist/x86_64-osx/Cabal-2.2.0.1/build/PostgREST/autogen/cabal_macros.h"'; specified path differs in case from file name on disk [-Werror,-Wnonportable-include-path]
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
".stack-work/dist/x86_64-osx/Cabal-2.2.0.1/build/PostgREST/autogen/cabal_macros.h"
* Update stack.yaml to lts-12.26 (GHC 8.4.4) with simplified nix config
(On GHC 8.6.5, profiled build was exiting with:
-- While building package postgrest-5.2.0 using:
/home/circleci/.stack/setup-exe-cache/x86_64-linux/Cabal-simple_mPHDZzAJ_2.4.0.1_ghc-8.6.5 --builddir=.stack-work/dist/x86_64-linux/Cabal-2.4.0.1 build lib:postgrest exe:postgrest --ghc-options " -ddump-hi -ddump-to-file -fdiagnostics-color=always"
Process exited with code: ExitFailure (-9) (THIS MAY INDICATE OUT OF MEMORY)
Exited with code 1)
* stack.yaml: Add postgresql to nix packages
It was detected that Aeson encoding had high memory usage when
the json payload was large, around x60 the payload size.
With this change we get around x10 payload size memory usage.
The encodeUtf8(when doing a Text -> ByteString with `toS`) function
on a large payload also contributed to the high memory usage.
Main idea to reduce the memory usage was to let the ByteString coming
from the request body go to the database unchanged.
