- Disabled by default.
- Enabled with flags: `enableInstrumentation` and `enableLightweightStepTracing'
The instrumentation:
- tracks #pushes and maximum-depth for the `kont`-stack and the `env`ironment
- counts #steps
- classifies each step:
- for `CtrlExpr`, what kind of expression is it?
- for `CtrlValue`, what kind of continuation is at the head of the `kont`-stack?
- the information is printed when the machine finalizes
The lightweight tracing shows an abbreviated output line per step:
- the control expression / value
- the size of the environment
- the depth of the `kont`-stack
changelog_begin
changelog_end
This is temporary. It looks like the macOS nodes are dead; @nycnewman is
looking into it, but in case he doesn't fix it in time, at least we
have a backup plan so we're not completely blocked on Monday.
CHANGELOG_BEGIN
CHANGELOG_END
add default machine capability
We semi-regularly need to do work that has the potential to disrupt a
machine's local cache, rendering it broken for other streams of work.
This can include upgrading nix, upgrading Bazel, debugging caching
issues, or anything related to Windows.
Right now we do not have any good solution for these situations. We can
either not do those streams of work, or we can proceed with them and
just accept that all other builds may get affected depending on which
machine they get assigned to. Debugging broken nodes is particularly
tricky as we do not have any way to force a build to run on a given
node.
This PR aims at providing a better alternative by (ab)using an Azure
Pipelines feature called
[capabilities](https://docs.microsoft.com/en-us/azure/devops/pipelines/agents/agents?view=azure-devops&tabs=browser#capabilities).
The idea behind capabilities is that you assign a set of tags to a
machine, and then a job can express its
[demands](https://docs.microsoft.com/en-us/azure/devops/pipelines/process/demands?view=azure-devops&tabs=yaml),
i.e. specify a set of tags machines need to have in order to run it.
Support for this is fairly badly documented. We can gather from the
documentation that a job can specify two things about a capability
(through its `demands`): that a given tag exists, and that a given tag
has an exact specified value. In particular, a job cannot specify that a
capability should _not_ be present, meaning we cannot rely on, say,
adding a "broken" tag to broken machines.
Documentation on how to set capabilities for an agent is basically
nonexistent, but [looking at the
code](https://github.com/microsoft/azure-pipelines-agent/blob/master/src/Microsoft.VisualStudio.Services.Agent/Capabilities/UserCapabilitiesProvider.cs)
indicates that they can be set by using a simple `key=value`-formatted
text file, provided we can find the right place to put this file.
This PR adds this file to our Linux, macOS and Windows node init scripts
to define an `assignment` capability and adds a demand for a `default`
value on each job. From then on, when we hit a case where we want a PR
to run on a specific node, and to prevent other PRs from running on that
node, we can manually override the capability from the Azure UI and
update the demand in the relevant YAML file in the PR.
CHANGELOG_BEGIN
CHANGELOG_END
* postgresql-testing: Store the JDBC URL separately.
* postgresql-testing: Expose the username and password.
* postgresql-testing: Get the caller to create the database.
And make sure it's a random one, not "test".
CHANGELOG_BEGIN
CHANGELOG_END
* postgresql-testing: Only store the JDBC URL for tests.
Less mutable state, innit.
* postgresql-testing: Capture the individual JDBC URL parameters.
* Bazel: Fix PostgreSQL binary paths.
* postgresql-testing: Just recreate the database in PostgresAroundEach.
There's no need to restart the process with a different data directory.
This means we can remove the dependency on `silencer-lib`, which means
that users don't have to upgrade to Scala v2.12.11.
CHANGELOG_BEGIN
- [Scala Bindings]: We no longer require users to upgrade to Scala
v2.12.11.
CHANGELOG_END
I’m not entirely sure why people end up with poisoned nix
caches (seems harder than poisoning your Bazel cache) but a few people
have managed apparently so this should hopefully help them recover.
changelog_begin
changelog_end
We add a new traversels to GenTransaction that visit all the nodes in execution order.
Exercise nodes are visited twice: when execution reaches them and when execution leaves their body.
CHANGELOG_BEGIN
CHANGELOG_END
This PR changes the notify_user job to not run when the job has been
canceled, which happens mostly when we push new code.
Not sure how I failed to see the `canceled` function in the past, but
this does seem to do exactly what we want.
CHANGELOG_BEGIN
CHANGELOG_END
* Use a table specifically for transaction trees witnesses
This allows to not re-compute those every time using the flat transaction
witnesses and the complement.
Furthermore, witness tables indexing is redefined to a single primary key
covering both event identifier and witness party instead of two
separated indexes. This has proven to be a relevant improvement for
performance.
Fixes#5883
changelog_begin
[Ledger API Server] Significant improvements in serving transaction trees. See #5883.
changelog_end
* Fix migration SHA
Fixes#5786.
CHANGELOG_BEGIN
[DAML Standard Lib] ``DA.Text.splitOn`` will now correctly handle the
case where the separator appears at the end but should not be matched,
as in ``splitOn "aa" "aaa" == ["", "a"]`` (previously this was
erroneously returning ``["", "a", ""]``). See #5786 for more details.
CHANGELOG_END
- Have a special continuation `KFinished` which is pushed on the initially empty `kontStack`.
- `KFinished.execute` throws `SpeedyHungry` when it has the `SResultFinalValue`.
- This unifies the behaviour w.r.t other kinds of `SResult`.
- The `kontStack` is never empty during evaluation.
- So we no longer require an`isFinal` test, and so the inner loop is a little tighter.
The performance improvement here is marginal (1% maybe). But the clarity is better. Also, it's a nice invariant that there is always at least one continuation on the stack. An invariant which will be useful in a change planned to improve handling of `KPop` and `KLocation`.
changelog_begin
changelog_end
This PR separates the "last known valid perf test" commit from the
"baseline speedy implementation" commit. It is important for the perf
test to be meaningful that the changes between those two commits are
benign, say minor API adjustments, so that the perf measurement remains
meaningful.
This also adds a check on merging to master that tells Slack if the perf
test has changed and the `test_sha` file needs updating. The Slack
message is conditional on the current commit to avoid excessive noise.
CHANGELOG_BEGIN
CHANGELOG_END
* ledger-on-sql: Use dedicated database execution contexts.
This may or may not stop us from monopolizing API server threads.
CHANGELOG_BEGIN
CHANGELOG_END
* ledger-on-sql: Name arguments to `Database` and `UninitializedDatabase`.
bazel configuration does two things:
It modifies .bazelrc.local and it writes a temp file.
I’ve run `git clean` on a PR. This caused the temp file to be
removed. However `.bazelrc.local` stayed since it is in
`.gitignore`. This meant that the next time the formatting check ran
the `.bazelrc.local` pointed to the temp file but the temp file was no
longer there.
changelog_begin
changelog_end
* Simplify and clarify the public interface to Speedy.
- Remove `isFinal`. A client just uses `run()`.
- Remove `toSValue`. The value in available in `SResultFinalValue(v: SValue)`.
- A client never directly access the `.ctrl` (or `.returnValue`) components.
- A client may use `setExpressionToEvaluate(expr)` to evaluate a new expression on an existing machine.
changelog_begin
changelog_end
* remove while loop which executes just once
* avoid unnecessary mutation when running speedy
While the tests are exclusive and I’ve kept them exclusive for
resource usage, the hardcoded port numbers can still cause issues if
for whatever reason you have a postgres instance running outside of
Bazel. This could happen because you simply have a local postgres
instance for testing other things or, what seems to have happened on
CI, due to a leftover postgres instance from a previous test.
While we should do something about the latter, this at least gets us
into a situation where the node isn’t completely broken at this point.
changelog_begin
changelog_end
As mentioned in the comment, this is required to get DNS requests to
work. This is more important than one might realize at first:
`daml start` tries to make an HTTP request to localhost:7500 to wait
for Navigator to start up. However, in our docker image, these
requests currently fail completely since they fail to resolve
`localhost`. This stops all following steps, in particular,
`init-script` and the JSON API from starting up.
changelog_begin
changelog_end
This should fix the following issue that we see constantly on CI:
```
ld: warning: object file _ was built for newer OSX version (10.15) than being linked (10.14)
```
The issue was that the CC toolchain was not fully used in
haskell_cabal_package. --with-gcc (which is really --with-cc) only
applies when Cabal is calling the C compiler. However, in most cases
it is actually GHC itself which calls the C compiler. To make sure
that the right compiler is used in those cases, we have to pass
`-pgmc` and friends to GHC. This matches what rules_haskell does for
non cabal targets.
changelog_begin
changelog_end
Remove the `Ctrl` trait and separate `Machine.ctrl: Ctrl` into `Machine.ctrl: SExpr` and `Machine.returnValue: SValue` instead. This allows for avoiding dynamic dispatch on `ctrl` and instead allows for checking a pointer for `null` to decide if we have an expression that needs further break-down or a return value ready to be passed to the next continuation.
To make this check really only a pointer comparison we also needed to remove the abomination of "fully applied partially applied primitives". In order to achieve this, we check whether a PAP will be fully applied afterward when applying the last argument.
On the `collect-authority` benchmark, this increases throughput by around 13%, on another more computation heave benchmark by about 21%.
`collect-authority` benchmark on `master`:
```
Result "com.daml.lf.speedy.perf.CollectAuthority.bench":
112.361 ±(99.9%) 1.965 ms/op [Average]
(min, avg, max) = (107.047, 112.361, 120.745), stdev = 3.493
CI (99.9%): [110.396, 114.326] (assumes normal distribution)
```
`collect-authority` benchmark on this branch:
```
Result "com.daml.lf.speedy.perf.CollectAuthority.bench":
98.196 ±(99.9%) 1.933 ms/op [Average]
(min, avg, max) = (91.580, 98.196, 105.478), stdev = 3.436
CI (99.9%): [96.263, 100.129] (assumes normal distribution)
```
computation heavy benchmark on master
```
Result "com.daml.lf.speedy.perf.CollectAuthority.bench":
44.030 ±(99.9%) 0.742 ms/op [Average]
(min, avg, max) = (42.124, 44.030, 46.781), stdev = 1.319
CI (99.9%): [43.289, 44.772] (assumes normal distribution)
```
computation heavy benchmark on this branch:
```
Result "com.daml.lf.speedy.perf.CollectAuthority.bench":
36.222 ±(99.9%) 0.580 ms/op [Average]
(min, avg, max) = (34.897, 36.222, 39.787), stdev = 1.031
CI (99.9%): [35.643, 36.802] (assumes normal distribution)
```
changelog_begin
changelog_end
* DAML-SCRIPT: cleanup to prepare #5811
* a bit more.
CHANGELOG_BEGIN
CHANGELOG_END
* Address Moritz's review
* Update daml-script/runner/src/main/scala/com/digitalasset/daml/lf/engine/script/Runner.scala
Co-authored-by: Martin Huschenbett <martin.huschenbett@posteo.me>
Co-authored-by: Martin Huschenbett <martin.huschenbett@posteo.me>
* Use triggerId field in trigger start response
* Use triggerId field for stop trigger result
* Fix indentation and make yields consistent
* Use pair constructor for JsObject instead of Map
* Use triggerIds field in list triggers response
changelog_begin
changelog_end
This PR adds a simple daily job that runs the performance test on a
chosen "baseline" commit and then runs the same benchmark on latest
master. This should allow us to track overall performance improvements.
CHANGELOG_BEGIN
CHANGELOG_END
* Implement version name shortening for Windows
CHANGELOG_BEGIN
CHANGELOG_END
* Temporarily disable `--quick`.
* Revert "Temporarily disable `--quick`."
This reverts commit 025e6c2ddd.
Co-authored-by: Andreas Herrmann <andreas.herrmann@tweag.io>
This changes the code for starting postgres to print the log if
`pg_ctl start` fails.
In addition to that it marks the migration tests exclusive since they
rely on this code which has a hardcoded port number.
I tested this on CI by temporarily disabling --quick.
changelog_begin
changelog_end
