## Changes
This PR adds the `cmd-exec-id` field to the user agent. This allows us
to correlate multiple HTTP requests made from the CLI.
### Why Not Use HTTP traceparent?
We considered using the traceparent header in HTTP as an alternative,
but it's not a good fit for our use case. Here's why:
1. Purpose of traceparent: It's designed to trace a single HTTP request
across a distributed system as it moves through subsystems and proxies.
2. Our requirement: We need to trace multiple HTTP requests made during
a single command execution in the CLI.
For more details about how traceparent itself works and how it's used in
the Go SDK, see
https://github.com/databricks/databricks-sdk-go/pull/914.
## Tests
Unit test
## Changes
We were not using the readers and writers set in the test fixtures in
the progress logger. This PR fixes that. It also modifies
`TestAccAbortBind`, which was implicitly relying on the bug.
I encountered this bug while working on
https://github.com/databricks/cli/pull/1672.
## Tests
Manually.
From non-tty:
```
Error: failed to bind the resource, err: This bind operation requires user confirmation, but the current console does not support prompting. Please specify --auto-approve if you would like to skip prompts and proceed.
```
From tty, bind works as expected.
```
Confirm import changes? Changes will be remotely applied only after running 'bundle deploy'. [y/n]: y
Updating deployment state...
Successfully bound databricks_pipeline with an id '9d2dedbb-f522-4503-96ba-4bc4d5bfa77d'. Run 'bundle deploy' to deploy changes to your workspace
```
## Changes
Add regression tests for https://github.com/databricks/cli/issues/1563
We test 2 code paths:
- if there is an error, we can print to stderr
- if there is a valid output, we can print to stdout
We should also consider adding black-box tests that will run the CLI
binary as a black box and inspect its output to stderr/stdout.
## Tests
Unit tests
## Changes
This snuck into #1532 right before merging. The result is that error
output is no longer logged. This includes actual execution errors as
well as help output if arguments or flags are incorrectly specified.
We don't have test coverage for the `root.Execute` function. This is to
be fixed later.
## Tests
Manually confirmed we observe error output again.
## Changes
This combination of changes allows pretty-printing errors happening
during the "load" and "init" phases, including their locations.
Move to render code into a separate module dedicated to rendering
`diag.Diagnostics` in a human-readable format. This will be used for the
`bundle deploy` command.
Preserve the "bundle" value if an error occurs in mutators. Rewrite the
go templates to handle the case when the bundle isn't yet loaded if an
error occurs during loading, that is possible now.
Improve rendering for errors and warnings:
- don't render empty locations
- render "details" for errors if they exist
Add `root.ErrAlreadyPrinted` indicating that the error was already
printed, and the CLI entry point shouldn't print it again.
## Tests
Add tests for output, that are especially handy to detect extra newlines
## Changes
Currently, there are a number of issues with the non-happy-path flows
for token refresh in the CLI.
If the token refresh fails, the raw error message is presented to the
user, as seen below. This message is very difficult for users to
interpret and doesn't give any clear direction on how to resolve this
issue.
```
Error: token refresh: Post "https://adb-<WSID>.azuredatabricks.net/oidc/v1/token": http 400: {"error":"invalid_request","error_description":"Refresh token is invalid"}
```
When logging in again, I've noticed that the timeout for logging in is
very short, only 45 seconds. If a user is using a password manager and
needs to login to that first, or needs to do MFA, 45 seconds may not be
enough time. to an account-level profile, it is quite frustrating for
users to need to re-enter account ID information when that information
is already stored in the user's `.databrickscfg` file.
This PR tackles these two issues. First, the presentation of error
messages from `databricks auth token` is improved substantially by
converting the `error` into a human-readable message. When the refresh
token is invalid, it will present a command for the user to run to
reauthenticate. If the token fetching failed for some other reason, that
reason will be presented in a nice way, providing front-line debugging
steps and ultimately redirecting users to file a ticket at this repo if
they can't resolve the issue themselves. After this PR, the new error
message is:
```
Error: a new access token could not be retrieved because the refresh token is invalid. To reauthenticate, run `.databricks/databricks auth login --host https://adb-<WSID>.azuredatabricks.net`
```
To improve the login flow, this PR modifies `databricks auth login` to
auto-complete the account ID from the profile when present.
Additionally, it increases the login timeout from 45 seconds to 1 hour
to give the user sufficient time to login as needed.
To test this change, I needed to refactor some components of the CLI
around profile management, the token cache, and the API client used to
fetch OAuth tokens. These are now settable in the context, and a
demonstration of how they can be set and used is found in
`auth_test.go`.
Separately, this also demonstrates a sort-of integration test of the CLI
by executing the Cobra command for `databricks auth token` from tests,
which may be useful for testing other end-to-end functionality in the
CLI. In particular, I believe this is necessary in order to set flag
values (like the `--profile` flag in this case) for use in testing.
## Tests
Unit tests cover the unhappy and happy paths using the mocked API
client, token cache, and profiler.
Manually tested
---------
Co-authored-by: Pieter Noordhuis <pieter.noordhuis@databricks.com>
## Changes
The function signature of Cobra's `PreRunE` function has an `error`
return value. We'd like to start returning `diag.Diagnostics` after
loading a bundle, so this is incompatible. This change modifies all
usage of `PreRunE` to load a bundle to inline function calls in the
command's `RunE` function.
## Tests
* Unit tests pass.
* Integration tests pass.
## Changes
Prior to this change, the bundle configuration entry point was loaded
from the function `bundle.Load`. Other configuration files were only
loaded once the caller applied the first set of mutators. This
separation was unnecessary and not ideal in light of gathering
diagnostics while loading _any_ configuration file, not just the ones
from the includes.
This change:
* Updates `bundle.Load` to only verify that the specified path is a
valid bundle root.
* Moves mutators that perform loading to `bundle/config/loader`.
* Adds a "load" phase that takes the place of applying
`DefaultMutators`.
Follow ups:
* Rename `bundle.Load` -> `bundle.Find` (because it no longer performs
loading)
This change depends on #1316 and #1317.
## Tests
Tests pass.
## Changes
This diagnostics type allows us to capture multiple warnings as well as
errors in the return value. This is a preparation for returning
additional warnings from mutators in case we detect non-fatal problems.
* All return statements that previously returned an error now return
`diag.FromErr`
* All return statements that previously returned `fmt.Errorf` now return
`diag.Errorf`
* All `err != nil` checks now use `diags.HasError()` or `diags.Error()`
## Tests
* Existing tests pass.
* I confirmed no call site under `./bundle` or `./cmd/bundle` uses
`errors.Is` on the return value from mutators. This is relevant because
we cannot wrap errors with `%w` when calling `diag.Errorf` (like
`fmt.Errorf`; context in https://github.com/golang/go/issues/47641).
## Changes
Currently, when the CLI run a list API call (like list jobs), it uses
the `List*All` methods from the SDK, which list all resources in the
collection. This is very slow for large collections: if you need to list
all jobs from a workspace that has 10,000+ jobs, you'll be waiting for
at least 100 RPCs to complete before seeing any output.
Instead of using List*All() methods, the SDK recently added an iterator
data structure that allows traversing the collection without needing to
completely list it first. New pages are fetched lazily if the next
requested item belongs to the next page. Using the List() methods that
return these iterators, the CLI can proactively print out some of the
response before the complete collection has been fetched.
This involves a pretty major rewrite of the rendering logic in `cmdio`.
The idea there is to define custom rendering logic based on the type of
the provided resource. There are three renderer interfaces:
1. textRenderer: supports printing something in a textual format (i.e.
not JSON, and not templated).
2. jsonRenderer: supports printing something in a pretty-printed JSON
format.
3. templateRenderer: supports printing something using a text template.
There are also three renderer implementations:
1. readerRenderer: supports printing a reader. This only implements the
textRenderer interface.
2. iteratorRenderer: supports printing a `listing.Iterator` from the Go
SDK. This implements jsonRenderer and templateRenderer, buffering 20
resources at a time before writing them to the output.
3. defaultRenderer: supports printing arbitrary resources (the previous
implementation).
Callers will either use `cmdio.Render()` for rendering individual
resources or `io.Reader` or `cmdio.RenderIterator()` for rendering an
iterator. This separate method is needed to safely be able to match on
the type of the iterator, since Go does not allow runtime type matches
on generic types with an existential type parameter.
One other change that needs to happen is to split the templates used for
text representation of list resources into a header template and a row
template. The template is now executed multiple times for List API
calls, but the header should only be printed once. To support this, I
have added `headerTemplate` to `cmdIO`, and I have also changed
`RenderWithTemplate` to include a `headerTemplate` parameter everywhere.
## Tests
- [x] Unit tests for text rendering logic
- [x] Unit test for reflection-based iterator construction.
---------
Co-authored-by: Andrew Nester <andrew.nester@databricks.com>
## Changes
This is a fundamental change to how we load and process bundle
configuration. We now depend on the configuration being represented as a
`dyn.Value`. This representation is functionally equivalent to Go's
`any` (it is variadic) and allows us to capture metadata associated with
a value, such as where it was defined (e.g. file, line, and column). It
also allows us to represent Go's zero values properly (e.g. empty
string, integer equal to 0, or boolean false).
Using this representation allows us to let the configuration model
deviate from the typed structure we have been relying on so far
(`config.Root`). We need to deviate from these types when using
variables for fields that are not a string themselves. For example,
using `${var.num_workers}` for an integer `workers` field was impossible
until now (though not implemented in this change).
The loader for a `dyn.Value` includes functionality to capture any and
all type mismatches between the user-defined configuration and the
expected types. These mismatches can be surfaced as validation errors in
future PRs.
Given that many mutators expect the typed struct to be the source of
truth, this change converts between the dynamic representation and the
typed representation on mutator entry and exit. Existing mutators can
continue to modify the typed representation and these modifications are
reflected in the dynamic representation (see `MarkMutatorEntry` and
`MarkMutatorExit` in `bundle/config/root.go`).
Required changes included in this change:
* The existing interpolation package is removed in favor of
`libs/dyn/dynvar`.
* Functionality to merge job clusters, job tasks, and pipeline clusters
are now all broken out into their own mutators.
To be implemented later:
* Allow variable references for non-string types.
* Surface diagnostics about the configuration provided by the user in
the validation output.
* Some mutators use a resource's configuration file path to resolve
related relative paths. These depend on `bundle/config/paths.Path` being
set and populated through `ConfigureConfigFilePath`. Instead, they
should interact with the dynamically typed configuration directly. Doing
this also unlocks being able to differentiate different base paths used
within a job (e.g. a task override with a relative path defined in a
directory other than the base job).
## Tests
* Existing unit tests pass (some have been modified to accommodate)
* Integration tests pass
## Changes
The JSON logger is excellent as a machine-readable logger with lots of
metadata, but the resulting logs are difficult to read:
<img width="1601" alt="Image_from_Databricks"
src="https://github.com/databricks/cli/assets/1850319/76aa852f-756f-4e0a-bc00-3a6e3224296a">
Currently, we only use the friendly log printer when run from a TTY.
This PR removes that restriction, so logs will be pretty-printed by
default, regardless of TTY or not. If a user needs machine-readable
logs, they can still use `--log-format JSON`.
## Tests
Manual test: `databricks current-user me --debug | cat` uses the
pretty-printing logger.
## Changes
Allow account client auth with environment variables when no
.databrickscfg file present
Makes the behaviour to be in line with WorkspaceClient auth.
## Tests
Added regression test
## Changes
It wasn't working because it deferred to the regular `slog.TextHandler`
for the `WithAttr` and `WithGroup` functions. Both of these functions
don't mutate the handler but return a new one. When the top-level logger
called one of these, log records in that context used the standard
handler instead of ours.
To implement tracking of attributes and groups, I followed the guide at
https://github.com/golang/example/blob/master/slog-handler-guide/README.md
for writing custom handlers.
## Tests
The new tests demonstrate formatting through `t.Log` and look good.
## Changes
We didn't return the error upon creating a workspace or account client.
If there is an error, it must always propagate up the stack. The result
of this bug was that we were setting a `nil` account or workspace
client, which in turn caused SIGSEGVs.
Fixes#913.
## Tests
Manually confirmed this fixes the linked issue. The CLI now correctly
returns an error when the client cannot be constructed.
The issue was reproducible using a `.databrickscfg` with a single,
incorrectly configured profile.
## Changes
`os.Getenv(..)` is not friendly with `libs/env`. This PR makes the
relevant changes to places where we need to read user home directory.
## Tests
Mainly done in https://github.com/databricks/cli/pull/914
## Changes
This will help differentiate multiple cli commands that write to the
same log file.
Noticed that the root module wasn't using the common log utilities,
refactored it to avoid missing log arguments.
Relevant PR on the databricks vscode extension side:
https://github.com/databricks/databricks-vscode/pull/923
## Tests
Tested manually for sdk and cli loggers
## Changes
If a bundle configuration specifies a workspace host, and the user
specifies a profile to use, we perform a check to confirm that the
workspace host in the bundle configuration and the workspace host from
the profile are identical. If they are not, we return an error. The
check was introduced in #571.
Previously, the code included an assumption that the client
configuration was already loaded from the environment prior to
performing the check. This was not the case, and as such if the user
intended to use a non-default path to `.databrickscfg`, this path was
not used when performing the check.
The fix does the following:
* Resolve the configuration prior to performing the check.
* Don't treat the configuration file not existing as an error.
* Add unit tests.
Fixes#884.
## Tests
Unit tests and manual confirmation.
## Changes
This is used for the sync command, where we need to ensure that a bundle
configuration never taints the authentication setup as prepared in the
environment (by our VS Code extension). Once the VS Code extension fully
builds on bundles, we can remove this check again.
## Tests
Manually confirmed that calling `databricks sync` from a bundle
directory no longer picks up its authentication configuration.
## Changes
The first stab at this was added in #837 but only included the
`NoPrompt` check in `MustAccountClient`. I renamed it to `SkipPrompt`
(in preparation for another option that skips bundle load) and made it
work for `MustWorkspaceClient` as well.
## Tests
Manually confirmed that the completion hook no longer prompts for a
profile (when called directly with `databricks __complete`).
## Changes
Fixes#836
## Tests
Manually running `sync` command with and without the flag
Integration tests pass as well
```
--- PASS: TestAccSyncFullFileSync (13.38s)
PASS
coverage: 39.1% of statements in ./...
ok github.com/databricks/cli/internal 14.148s coverage: 39.1% of statements in ./...
--- PASS: TestAccSyncIncrementalFileSync (11.38s)
PASS
coverage: 39.1% of statements in ./...
ok github.com/databricks/cli/internal 11.674s coverage: 39.1% of statements in ./...
```
## Changes
If the caller running the test has one or more environment variables
that are used in the test already set, they can interfere and make tests
fail.
## Tests
Ran tests in `./cmd/root` with Databricks related environment variables
set.
## Changes
The previous implementation ran the risk of infinite looping for the
account client due to a mismatch in determining what constitutes an
account client between the CLI and SDK (see
[here](83443bae8d/libs/databrickscfg/profiles.go (L61))
and
[here](0fdc5165e5/config/config.go (L160))).
Ultimately, this code must never infinite loop. If a user is prompted
and selects a profile that cannot be used, they should receive that
feedback immediately and try again, instead of being prompted again.
Related to #726.
## Tests
<!-- How is this tested? -->
## Changes
There are a couple places throughout the code base where interaction
with environment variables takes place. Moreover, more than one of these
would try to read a value from more than one environment variable as
fallback (for backwards compatibility). This change consolidates those
accesses.
The majority of diffs in this change are mechanical (i.e. add an
argument or replace a call).
This change:
* Moves common environment variable lookups for bundles to
`bundles/env`.
* Adds a `libs/env` package that wraps `os.LookupEnv` and `os.Getenv`
and allows for overrides to take place in a `context.Context`. By
scoping overrides to a `context.Context` we can avoid `t.Setenv` in
testing and unlock parallel test execution for integration tests.
* Updates call sites to pass through a `context.Context` where needed.
* For bundles, introduces `DATABRICKS_BUNDLE_ROOT` as new primary
variable instead of `BUNDLE_ROOT`. This was the last environment
variable that did not use the `DATABRICKS_` prefix.
## Tests
Unit tests pass.
## Changes
This reduces the latency of every workspace command by the duration of a
single API call to retrieve the current user (which can take up to a
full second).
Note: the better place to verify that a request can be authenticated is
the SDK itself.
## Tests
* Unit test to confirm an the empty `*http.Request` can be constructed
* Manually confirmed that the additional API call no longer happens
## Changes
This pull request extends the templating support in preparation of a
new, default template (WIP, https://github.com/databricks/cli/pull/686):
* builtin templates that can be initialized using e.g. `databricks
bundle init default-python`
* builtin templates are embedded into the executable using go's `embed`
functionality, making sure they're co-versioned with the CLI
* new helpers to get the workspace name, current user name, etc. help
craft a complete template
* (not enabled yet) when the user types `databricks bundle init` they
can interactively select the `default-python` template
And makes two tangentially related changes:
* IsServicePrincipal now uses the "users" API rather than the
"principals" API, since the latter is too slow for our purposes.
* mode: prod no longer requires the 'target.prod.git' setting. It's hard
to set that from a template. (Pieter is planning an overhaul of warnings
support; this would be one of the first warnings we show.)
The actual `default-python` template is maintained in a separate PR:
https://github.com/databricks/cli/pull/686
## Tests
Unit tests, manual testing
## Changes
Renamed Environments to Targets in bundle.yml.
The change is backward-compatible and customers can continue to use
`environments` in the time being.
## Tests
Added tests which checks that both `environments` and `targets` sections
in bundle.yml works correctly
## Changes
#629 introduced a change to autopopulate the host from .databrickscfg if
the user is logging back into a host they were previously using. This
did not respect the DATABRICKS_CONFIG_FILE env variable, causing the
flow to stop working for users with no .databrickscfg file in their home
directory.
This PR refactors all config file loading to go through one interface,
`databrickscfg.GetDatabricksCfg()`, and an auxiliary
`databrickscfg.GetDatabricksCfgPath()` to get the configured file path.
Closes#655.
## Tests
```
$ databricks auth login --profile abc
Error: open /Users/miles/.databrickscfg: no such file or directory
$ ./cli auth login --profile abc
Error: cannot load Databricks config file: open /Users/miles/.databrickscfg: no such file or directory
$ DATABRICKS_CONFIG_FILE=~/.databrickscfg.bak ./cli auth login --profile abc
Databricks Host: https://asdf
```
## Changes
This removes the remaining dependency on global state and unblocks work
to parallelize integration tests. As is, we can already uncomment an
integration test that had to be skipped because of other tests tainting
global state. This is no longer an issue.
Also see #595 and #606.
## Tests
* Unit and integration tests pass.
* Manually confirmed the help output is the same.
## Changes
This change is another step towards a CLI without globals. Also see #595.
The flags for the root command are now encapsulated in struct types.
## Tests
Unit tests pass.
## Changes
Generated commands relied on global variables for flags and request
payloads. This is difficult to test if a sequence of tests tries to run
the same command with various arguments because the global state causes
test interference. Moreover, it is impossible to run tests in parallel.
This change modifies the approach and turns every command group and
command itself into a function that returns a `*cobra.Command`. All
flags and request payloads are variables scoped to the command's
initialization function. This means it is possible to construct
independent copies of the CLI structure and fixes the test isolation
issue.
The scope of this change is only the generated commands. The other
commands will be changed accordingly in subsequent changes.
## Tests
Unit and integration tests pass.
## Changes
Correctly use --profile flag passed for all bundle commands.
Also adds a validation that if bundle configured host mismatches
provided profile, it throws an error.
Co-authored-by: Pieter Noordhuis <pieter.noordhuis@databricks.com>
## Changes
Currently, `databricks --profile <TAB>` autocompletes with the shell
default behavior, listing files in the local directory. This is not a
great experience. Especially given that the suggested profile names for
accounts are so long, it can be cumbersome to type them out by hand.
This PR configures autocompletion for `--profile` to inspect the
profiles of ~/.databrickscfg.
One potential improvement is to filter the response based on whether the
command is known to be account-level or workspace-level.
## Tests
Manual test.
<img width="579" alt="Screenshot_11_07_2023__18_31"
src="https://github.com/databricks/cli/assets/1850319/d7a3acd0-2511-45ac-bd82-95567775c10a">
## Changes
This includes the following changes:
* Move profile loading code to libs/databrickscfg and add tests
* Update prompt label to reflect workspace/account profiles
* Start prompt in search mode by default
* Custom error if `~/.databrickscfg` doesn't exist
* Custom error if `~/.databrickscfg` doesn't contain profiles
* Use stderr for prompt so that stdout redirection works (e.g. with `jq` or `jless`)
## Tests
* New unit tests pass
* Manual tests for both workspace and account commands
* Search-by-default is really nice if you have many profiles
## Changes
Added support for `bundle.Seq`, simplified `Mutator.Apply` interface by
removing list of mutators from return values/
## Tests
1. Ran `cli bundle deploy` and interrupted it with Cmd + C mid execution
so lock is not released
2. Ran `cli bundle deploy` top make sure that CLI is not trying to
release lock when it fail to acquire it
```
andrew.nester@HFW9Y94129 multiples-tasks % cli bundle deploy
Starting upload of bundle files
Uploaded bundle files at /Users/andrew.nester@databricks.com/.bundle/simple-task/development/files!
^C
andrew.nester@HFW9Y94129 multiples-tasks % cli bundle deploy
Error: deploy lock acquired by andrew.nester@databricks.com at 2023-05-24 12:10:23.050343 +0200 CEST. Use --force to override
```
## Changes
With this PR, all of the command below print version and exit:
```
$ databricks -v
Databricks CLI v0.100.1-dev+4d3fa76
$ databricks --version
Databricks CLI v0.100.1-dev+4d3fa76
$ databricks version
Databricks CLI v0.100.1-dev+4d3fa76
```
## Tests
Added integration test for each flag or command.
shreyas-goenka
Andrew Nester
Gleb Kanterov
Pieter Noordhuis
Miles Yucht
Serge Smertin
Ilia Babanov
Lennart Kats (databricks)
