If a blob's reported size is incorrect, `archiveTar` may attempt to
write more bytes to an entry than it reported in that entry's header.
These changes provide a bit more context with the resulting error as
well as removing an unnecessary `LimitReader`.
Adds OpenTracing in the Pulumi engine and plugin + langhost subprocesses.
We currently create a single root span for any `Enging.plan` operation - which is a single `preview`, `update`, `destroy`, etc.
The only sub-spans we currently create are at gRPC boundaries, both on the client and server sides and on both the langhost and provider plugin interfaces.
We could extend this to include spans for any other semantically meaningful sections of compute inside the engine, though initial examples show we get pretty good granularity of coverage by focusing on the gRPC boundaries.
In the future, this should be easily extensible to HTTP boundaries and to track other bulky I/O like datastore read/writes once we hook up to the PPC and Pulumi Cloud.
We expose a `--trace <endpoint>` option to enable tracing on the CLI, which we will aim to thread through to subprocesses.
We currently support sending tracing data to a Zipkin-compatible endpoint. This has been validated with both Zipkin and Jaeger UIs.
We do not yet have any tracing inside the TypeScript side of the JS langhost RPC interface. There is not yet automatic gRPC OpenTracing instrumentation (though it looks like it's in progress now) - so we would need to manually create meaningful spans on that side of the interface.
We currently have a nasty issue with archive assets wherein they read
their entire contents into memory each time they are accessed (e.g. for
hashing or translation). This interacts badly with scenarios that
place large amounts of data in an archive: aside from limiting the size
of an archive the engine can handle, it also bloats the engine's memory
requirements. This appears to have caused issues when running the PPC in
AWS: evidence suggests that the very high peak memory requirements this
approach implies caused high swap traffic that impacted the service's
availability.
In order to fix this issue, these changes move archives onto a
streaming read model. In order to read an archive, a user:
- Opens the archive with `Archive.Open`. This returns an ArchiveReader.
- Iterates over its contents using `ArchiveReader.Next`. Each returned
blob must be read in full between successive calls to
`ArchiveReader.Next`. This requirement is essentially forced upon us
by the streaming nature of TAR archives.
- Closes the ArchiveReader with `ArchiveReader.Close`.
This model does not require that the complete contents of the archive or
any of its constituent files are in memory at any given time.
Fixes#325.
This code was swallowing an error incorrectly, rather than returning
it. As a result, certain commands would fail with assertions rather
than the intended error message (like trying to config without a stack).
Rework the polling code for `pulumi` when targeting hosted scenarios. (i.e. absorb https://github.com/pulumi/pulumi-service/pull/282.) We now return an `updateID` for update, preview, and destroy. And use bake that into the URL.
This PR also silently ignores 504 errors which are now returned from the Pulumi Service if the PPC request times out.
Combined, this should ameliorate some of the symptoms we see from https://github.com/pulumi/pulumi-ppc/issues/60. Though we'll continue to try to identify and fix the root cause.
We are not ready to start running integration tests against
Pulumi.com, but `pulumi` uses `PULUMI_API` to decide what backend to
target. Remove it when running `pulumi` in the integration tests so we
always get the local behavior.
pulumi/pulumi#471 tracks the work to actually be able to choose what
backend you want when an integration test runs.
This changes the overall flow to exit early when a failure occurs.
We will try to clean up if we believe to have gotten far enough to
actually have provisioned any resources.
In our existing code, we only use the input state for old and new
properties. This is incorrect and I'm astonished we've been flying
blind for so long here. Some resources require the output properties
from the prior operation in order to perform updates. Interestingly,
we did correclty use the full synthesized state during deletes.
I ran into this with the AWS Cloudfront Distribution resource,
which requires the etag from the prior operation in order to
successfully apply any subsequent operations.
We were previously calling configure on each package once per time it was mentioned in the config. We only need to call it once ever as we pass the full bag of relevent config through on that one call.
It's legal and possible for undefined properties to show up in
objects, since that's an idiomatic JavaScript way of initializing
missing properties. Instead of failing for these during deployment,
we should simply skip marshaling them to Terraform and let it do
its thing as usual. This came up during our customer workload.
Previously, we stored configuration information in the Pulumi.yaml
file. This was a change from the old model where configuration was
stored in a special section of the checkpoint file.
While doing things this way has some upsides with being able to flow
configuration changes with your source code (e.g. fixed values for a
production stack that version with the code) it caused some friction
for the local development scinerio. In this case, setting
configuration values would pend changes to Pulumi.yaml and if you
didn't want to publish these changes, you'd have to remember to remove
them before commiting. It also was problematic for our examples, where
it was not clear if we wanted to actually include values like
`aws:config:region` in our samples. Finally, we found that for our
own pulumi service, we'd have values that would differ across each
individual dev stack, and publishing these values to a global
Pulumi.yaml file would just be adding noise to things.
We now adopt a hybrid model, where by default configuration is stored
locally, in the workspace's settings per project. A new flag `--save`
tests commands to actual operate on the configuration information
stored in Pulumi.yaml.
With the following change, we have have four "slots" configuration
values can end up in:
1. In the Pulumi.yaml file, applies to all stacks
2. In the Pulumi.yaml file, applied to a specific stack
3. In the local workspace.json file, applied to all stacks
4. In the local workspace.json file, applied to a specific stack
When computing the configuration information for a stack, we apply
configuration in the above order, overriding values as we go
along.
We also invert the default behavior of the `pulumi config` commands so
they operate on a specific stack (i.e. how they did before
e3610989). If you want to apply configuration to all stacks, `--all`
can be passed to any configuration command.
This change introduces an abstraction for a `backend` which manages
the implementation of some CLI commands. As part of defining the
interface, we introduce a new local backend implementation that just
uses data local to the machine.
This will let us share argument parsing and some display information
between the local case and the pulumi.com case in the CLI. We can
continue to refine this interface over time (e.g. today we have the
implementation of the Destroy/Update/Preview actually writing output
but instead they should be returning strongly typed data that the CLI
knows how to display and is unified across Pulumi.com deploys and
local deploys).
But this is a good first step.
