Replace D-Bus with HTTP-based clients #1438

mvidner · 2024-07-04T12:41:03Z

Problem

https://trello.com/c/hvPtBtMD

When moving to the new HTTP-based architecture, we took some shortcuts. One of them was not using HTTP clients in the command-line interface. We need to adapt the following clients:

Solution

UsersClient - still needed, because the HTTP service uses it to talk to the Ruby backend. UsersHTTPClient added.
LocalizationClient replaced with LocalizationHTTPClient

The rest will be done in subsequent PRs.

Testing

Tested manually with various correct and incorrect configs via agama config show and agama config load. This is ripe for automation, even in this PR.
aaand: added tests that
- set up a trivial HTTP server (using httpmock - is the dependency size OK?),
- mocking the JSON data and
- asserting what the Store classes make out of it

Screenshots

If the fix affects the UI attach some screenshots here.

mvidner · 2024-07-24T08:54:20Z

One odd thing, OpenAPI seems to omit the /api prefix for /users/first, but the server logs include it

mvidner · 2024-07-24T09:02:37Z

@jreidinger do you want to have a quick look at this stage?

rust/agama-lib/src/base_http_client.rs

coveralls · 2024-07-29T12:46:33Z

coverage: 71.228% (+0.3%) from 70.945%
when pulling 3ea01bd on dbus-to-http
into 83ee1d3 on master.

mvidner · 2024-07-31T07:14:19Z

Argh, my flawless tests fail CI with

Error: NotAuthenticated
test localization::store::test::test_getting_l10n ... FAILED

it looks like I mock the call changed in this PR but the auth call still goes to the live service?

to enable - connecting to alternative servers, either production or test mocks - unauthenticated calls, either for the initial auth or test mocks This is needed for both #1438 and #1495 In the end, the API changes only by adding a `Default` trait implementation which does no authentication. Connecting to a mock server is achieved by assigning to the public `base_url` field.

enable debug prints in BaseHTTPClient

Problem: When there is an issue ("trying to use reserved username") it is reported as > Anyhow(Backend call failed with status 422 and text '["There is a conflict between the entered\nusername and an existing username.\nTry another one."]') (which is a wrapped `ServiceError::BackendError`) even though the CLI is ready to handle a `ServiceError::WrongUser` Solution: let BaseHTTPClient::put return a deserialized response but this commit still does not parse the JSON and does not convert the Err

put_void (a PUT that ignores the reponse body)

- remove debug prints - document `request_response` and use it - document `put_void` (which ignores response body, whereas `put` deserializes it)

- cargo fmt --all -- - remove unused UsersStore arg - agama-server does not depend on reqwest

so that the CLI client can use it too

as suggested by Josef

failure expected, no mocking yet

httpmock - It is what I found with DDG: rust mock http request No idea about alternatives, this one fits well my simple use case, but does bring a load of dependencies (how much?)

mchf · 2024-08-02T10:44:13Z

rust/agama-lib/src/users/store.rs

-        if !success {
-            return Err(ServiceError::WrongUser(issues));
-        }
+        self.users_client.set_first_user(&first_user).await?;


why is suddenly possible to ignore failed case?

The D-Bus client set_first_user returns a tuple (success,issues) in band, as zbus::Result<(bool, Vec<String>)>
https://github.com/openSUSE/agama/blob/b9a73b57fc71f6eb47b3f98154ec61a31f4d57d9/rust/agama-lib/src/users/client.rs#L84

But the HTTP client set_first_user returns a Result<(), ...> where all the errors are moved to the Err alternative, and the WrongUser is handled on that lower level:
https://github.com/openSUSE/agama/blob/3dca6b5ed6a30389b0023aea9a1ea031934a5859/rust/agama-lib/src/users/http_client.rs#L26

(and it is the ? in this line that propagates the Err)

rust/agama-server/src/users/web.rs

imobachgs

Good job! In general, the PR looks good, although there are a few comments.

Checking the BaseHTTPClient client again and how it is used, I am still not convinced about those "raw" methods (get_response, post_response, etc.). I am not even sure if we will need them. Additionally, they leak the Response from reqwest and I am not sure whether I like that.

Perhaps it is out of the scope of this PR. Let's see if we need them at some point.

rust/agama-lib/src/base_http_client.rs

rust/agama-lib/src/localization/http_client.rs

rust/agama-lib/src/localization/store.rs

imobachgs · 2024-08-02T11:22:34Z

rust/agama-lib/src/users/http_client.rs

+    /// Set the configuration for the first user
+    pub async fn set_first_user(&self, first_user: &FirstUser) -> Result<(), ServiceError> {
+        let result = self.client.put_void("/users/first", first_user).await;
+        if let Err(ServiceError::BackendError(422, ref issues_s)) = result {


np: perhaps we could use the http::status::StatusCode from the HTTP crate.

Yup, but then I can't use if let like this... that's fine, a match will look good too

rust/agama-lib/src/users/http_client.rs

mvidner · 2024-08-05T08:43:32Z

@imobachgs thanks for the review, yes I left the BaseHTTPClient API messy and I will go for a cleanup:

*_response will be private (questions::HTTPClient will need fixing for that but it will become much simpler)
foo will deserialize the response and foo_void will return nothing

Unused by default but small enough to be included anyway. Useful to trace what the httpmock server actually receives in tests: When a test misbehaves, turn on logging for it: ```rs env_logger::init(); ``` and also set the log level to see the requests and responses: ```console $ (cd rust; RUST_LOG=httpmock=debug cargo test) ```

imobachgs

It looks better now, although I am not convinced on the use case of those _void functions.

imobachgs · 2024-08-06T08:59:34Z

rust/agama-lib/src/localization/store.rs

@@ -142,9 +143,11 @@ mod test {
        };
        let result = store.store(&settings).await;

+        // main assertion


If we are not using result for anything else, what about just adding the question mark at the end of the previous line?

store.store(&settings).await?;

The resulting behavior is of course identical.
Writing it in this way means to draw your attention to the main assertion of the test.

Actually I was bothered by the test code being obscure that I even looked up Rust RSpec in the hope to write tests that better express the intent. Well, BDD.

Apart from the constructor, only these are left in the public API: ``` get delete_void post post_void put put_void patch patch_void ``` The fns with `_void` ignore the body of a successful response and return `Ok(())`. The other fns expect a JSON body and deserialize it, returning `Ok(T)`. Unsuccessful responses return `Err(ServiceError::BackendError(status_code, body_as_string))` Other fns were deleted or made private. `*_response`: replaced by a private helper `method_response` FIXME: adjust the doc comments once we agree this API makes sense

Previously I was running ```rs let http_mock = ...; let result = object.method_under_test(); http_mock.assert(); assert!(result.is_ok()); ``` But when something unexpectedly fails, the `result` will be `Err` and we want to see it. ```rs let http_mock = ...; let result = object.method_under_test(); // main assertion FIRST, and if it fails we get to see the `Err` result?; // only then make sure that the mock server got calls as expected http_mock.assert(); ```

cargo clippy --no-deps

imobachgs

It looks good to me but, please, add a changes entry.

imobachgs

LGTM. Thanks!

About the *_void methods, we agreed to review our HTTP API to be more consistent and, eventually, we could get rid of them. But it is out of the scope of this PR.

## Problem Problem description is in issue #1539 ## Solution Implement sd_notify for agama-web-service to notify systemd when web service is ready to serve requests. Improve systemd dependencies to better reflect dependencies. For agama-auto adapt dependencies to reflect that CLI switch from direct systemd communication to HTTP API. related PR is #1438 ## Testing - *Tested manually* ## Screenshots snippet from log before change ( truncated, but to demonstrate that agama-auto runs too soon): ``` Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Started Agama Installer Service. Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Started Agama Web Server. Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Started Agama automatic profile runner. Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Starting Postpone login prompt after the SSL fingerprint issue is generated... Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + export YAST_SKIP_XML_VALIDATION=1 Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + YAST_SKIP_XML_VALIDATION=1 Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + '[' -z '' ']' Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2691]: ++ awk -F agama.auto= '{sub(/ .*$/, "", $2); print $2}' Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + url= Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + '[' -z '' ']' Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + echo 'no autoinstallation profile' Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: no autoinstallation profile Aug 14 13:10:47 dhcp118.suse.cz agama-auto[2689]: + exit 0 Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: agama-auto.service: Deactivated successfully. Aug 14 13:10:47 dhcp118.suse.cz sshd-gen-keys-start[2683]: ssh-keygen: generating new host keys: RSA ECDSA ED25519 Aug 14 13:10:47 dhcp118.suse.cz sshd[2702]: Server listening on 0.0.0.0 port 22. Aug 14 13:10:47 dhcp118.suse.cz sshd[2702]: Server listening on :: port 22. Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Started OpenSSH Daemon. Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Starting Generate issue file for SSH host keys... Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: agama-ssh-issue.service: Deactivated successfully. Aug 14 13:10:47 dhcp118.suse.cz systemd[1]: Finished Generate issue file for SSH host keys. Aug 14 13:10:47 dhcp118.suse.cz agama-web-server[2688]: registering event source with poller: token=Token(94923250365696), interests=READABLE | WRITABLE Aug 14 13:10:47 dhcp118.suse.cz agama-web-server[2688]: Initializing Aug 14 13:10:47 dhcp118.suse.cz agama-web-server[2688]: Waiting for DATA or OK from server Aug 14 13:10:47 dhcp118.suse.cz agama-web-server[2688]: Received OK from server Aug 14 13:10:47 dhcp118.suse.cz agama-web-server[2688]: Waiting for Unix FD passing agreement from server Aug 14 13:10:47 dhcp118.suse.cz agama-web-server[2688]: Unix FD passing agreed by server ``` and now after change: ``` Aug 15 10:19:15 dhcp118.suse.cz systemd[1]: Started Agama Web Server. Aug 15 10:19:15 dhcp118.suse.cz agama-web-server[12467]: Starting Agama web server at :::80 Aug 15 10:19:15 dhcp118.suse.cz agama-web-server[12467]: Starting Agama web server at :::443 ... Aug 15 10:19:15 dhcp118.suse.cz agama-web-server[12467]: Waiting for message on the socket.. ... Aug 15 10:19:15 dhcp118.suse.cz agama-web-server[12467]: Listening for property changes on org.freedesktop.NetworkManager.Device... Aug 15 10:19:15 dhcp118.suse.cz systemd[1]: Started Agama automatic profile runner. Aug 15 10:19:15 dhcp118.suse.cz agama-web-server[12467]: Message received on the socket: Msg { type: Signal, sender: UniqueName(Str(Borrowed(":1.2"))), path: ObjectPath("/org/freedesktop/systemd1/unit/agama_2dauto_2eservice"), iface: InterfaceName(Str(Borrowed("org.freedesktop.DBus.Properties"))), member: MemberName(Str(Borrowed("PropertiesChanged"))), body: Signature("sa{sv}as") } Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + export YAST_SKIP_XML_VALIDATION=1 Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + YAST_SKIP_XML_VALIDATION=1 Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + '[' -z '' ']' ... Aug 15 10:19:15 dhcp118.suse.cz agama-web-server[12467]: Waiting for message on the socket.. Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12482]: ++ awk -F agama.auto= '{sub(/ .*$/, "", $2); print $2}' Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + url= Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + '[' -z '' ']' Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + echo 'no autoinstallation profile' Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: no autoinstallation profile Aug 15 10:19:15 dhcp118.suse.cz agama-auto[12479]: + exit 0 Aug 15 10:19:15 dhcp118.suse.cz systemd[1]: agama-auto.service: Deactivated successfully. ```

Prepare for releasing Agama 10· * #1263 * #1330 * #1407 * #1408 * #1410 * #1411 * #1412 * #1416 * #1417 * #1419 * #1420 * #1421 * #1422 * #1423 * #1424 * #1425 * #1428 * #1429 * #1430 * #1431 * #1432 * #1433 * #1436 * #1437 * #1438 * #1439 * #1440 * #1441 * #1443 * #1444 * #1445 * #1449 * #1450 * #1451 * #1452 * #1453 * #1454 * #1455 * #1456 * #1457 * #1459 * #1460 * #1462 * #1464 * #1465 * #1466 * #1467 * #1468 * #1469 * #1470 * #1471 * #1472 * #1473 * #1475 * #1476 * #1477 * #1478 * #1479 * #1480 * #1481 * #1482 * #1483 * #1484 * #1485 * #1486 * #1487 * #1488 * #1489 * #1491 * #1492 * #1493 * #1494 * #1496 * #1497 * #1498 * #1499 * #1500 * #1501 * #1502 * #1503 * #1504 * #1505 * #1506 * #1507 * #1508 * #1510 * #1511 * #1512 * #1513 * #1514 * #1515 * #1516 * #1517 * #1518 * #1519 * #1520 * #1522 * #1523 * #1524 * #1525 * #1526 * #1527 * #1528 * #1529 * #1530 * #1531 * #1532 * #1533 * #1534 * #1535 * #1536 * #1537 * #1540 * #1541 * #1543 * #1544 * #1545 * #1546 * #1547 * #1548 * #1549 * #1550 * #1552 * #1553 * #1554 * #1555 * #1556 * #1557 * #1558 * #1559 * #1560 * #1562 * #1563 * #1565 * #1566 * #1567 * #1568 * #1569 * #1570 * #1571 * #1572 * #1573 * #1574 * #1575 * #1576 * #1577 * #1578 * #1579 * #1580 * #1581 * #1583 * #1584 * #1585 * #1586 * #1587 * #1588 * #1589 * #1590 * #1591 * #1592 * #1593 * #1596 * #1597 * #1598 * #1600 * #1602 * #1605 * #1606 * #1607 * #1608 * #1610 * #1611 * #1612 * #1613 * #1614 * #1619 * #1620 * #1621

mvidner force-pushed the dbus-to-http branch 2 times, most recently from c49b025 to 5d0396e Compare July 24, 2024 08:49

mvidner force-pushed the dbus-to-http branch from 5d0396e to cac68d4 Compare July 26, 2024 11:04

mvidner marked this pull request as ready for review July 26, 2024 11:25

jreidinger reviewed Jul 26, 2024

View reviewed changes

rust/agama-lib/src/base_http_client.rs Show resolved Hide resolved

mvidner force-pushed the dbus-to-http branch from 398fb2e to 7e23e90 Compare July 29, 2024 12:41

mvidner mentioned this pull request Jul 31, 2024

BaseHTTPClient: better constructor API, Default #1517

Merged

mvidner added 18 commits August 1, 2024 16:53

run zypper --verbose to see where it gets stuck

756fc6a

UsersHttpClient

b771383

derive Debug for Question

e295016

enable debug prints in BaseHTTPClient

Move RootConfig, RootPatchSettings to a model module

7ea3c0a

set root password, set root ssh key (happy path)

fc3f848

UsersHttpClient::set_first_user use Result(<>), issues as Err

dd32fc5

put_void (a PUT that ignores the reponse body)

BaseHTTPClient cleanup

5abaf51

- remove debug prints - document `request_response` and use it - document `put_void` (which ignores response body, whereas `put` deserializes it)

cleanup

298cabc

- cargo fmt --all -- - remove unused UsersStore arg - agama-server does not depend on reqwest

hint for fixing a failed test run

e094efb

consistent spelling: UsersHTTPClient

3458c21

Split out agama_lib::localization::model::LocaleConfig

9c71690

so that the CLI client can use it too

use LocalizationHTTPClient

074cbb4

remove LocalizationClient (D-Bus), no longer used

6fcbf2f

chore: fix cargo doc links

e5cd7b4

BaseHTTPClient::unit_or_error factored out

e6ccc55

as suggested by Josef

test: LocalizationStore and LocalizationHTTPClient on a happy path

b8b3541

failure expected, no mocking yet

test: make it green by running a mock HTTP server

7369123

httpmock - It is what I found with DDG: rust mock http request No idea about alternatives, this one fits well my simple use case, but does bring a load of dependencies (how much?)

test: factor out setup code

8dad93c

mchf reviewed Aug 2, 2024

View reviewed changes

rust/agama-server/src/users/web.rs Outdated Show resolved Hide resolved

imobachgs reviewed Aug 2, 2024

View reviewed changes

mvidner added 2 commits August 2, 2024 21:19

remove unneeded async/await

3dca6b5

patch_root: pass the u32 from backend

5c649ab

mvidner added 2 commits August 5, 2024 16:25

Hint to the user when AuthToken::find fails

6da0409

imobachgs reviewed Aug 6, 2024

View reviewed changes

mvidner added 8 commits August 7, 2024 14:05

fix up mock for u32

2166a9b

debug: show how to peek into reqwest::Response bodies

ebe5210

questions::HTTPClient::new remove unneeded async/await

b7b75c7

test questions::HTTPClient::list_questions

80b82e5

test questions::HTTPClient::try_answer

d2bff01

test: questions:HTTPClient::create_question

b0a6a20

mvidner force-pushed the dbus-to-http branch from 3ea01bd to b0a6a20 Compare August 7, 2024 12:06

simplify new code as Clippy suggested

bab32b9

cargo clippy --no-deps

imobachgs reviewed Aug 8, 2024

View reviewed changes

changelog

4817cb0

imobachgs approved these changes Aug 9, 2024

View reviewed changes

mvidner merged commit c992adc into master Aug 9, 2024
1 check passed

mvidner deleted the dbus-to-http branch August 9, 2024 09:20

mvidner mentioned this pull request Aug 9, 2024

feat(rust): allow passing the API URL through the CLI #1495

Merged

jreidinger mentioned this pull request Aug 15, 2024

Fix systemd startup #1541

Merged

imobachgs mentioned this pull request Sep 20, 2024

Update to release 10 #1622

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Replace D-Bus with HTTP-based clients #1438

Replace D-Bus with HTTP-based clients #1438

mvidner commented Jul 4, 2024 •

edited

Loading

mvidner commented Jul 24, 2024

mvidner commented Jul 24, 2024

coveralls commented Jul 29, 2024 •

edited

Loading

mvidner commented Jul 31, 2024

mchf Aug 2, 2024

mvidner Aug 2, 2024

imobachgs left a comment

imobachgs Aug 2, 2024

mvidner Aug 5, 2024

mvidner commented Aug 5, 2024

imobachgs left a comment

imobachgs Aug 6, 2024

mvidner Aug 9, 2024

imobachgs left a comment

imobachgs left a comment

Replace D-Bus with HTTP-based clients #1438

Replace D-Bus with HTTP-based clients #1438

Conversation

mvidner commented Jul 4, 2024 • edited Loading

Problem

Solution

Testing

Screenshots

mvidner commented Jul 24, 2024

mvidner commented Jul 24, 2024

coveralls commented Jul 29, 2024 • edited Loading

mvidner commented Jul 31, 2024

mchf Aug 2, 2024

Choose a reason for hiding this comment

mvidner Aug 2, 2024

Choose a reason for hiding this comment

imobachgs left a comment

Choose a reason for hiding this comment

imobachgs Aug 2, 2024

Choose a reason for hiding this comment

mvidner Aug 5, 2024

Choose a reason for hiding this comment

mvidner commented Aug 5, 2024

imobachgs left a comment

Choose a reason for hiding this comment

imobachgs Aug 6, 2024

Choose a reason for hiding this comment

mvidner Aug 9, 2024

Choose a reason for hiding this comment

imobachgs left a comment

Choose a reason for hiding this comment

imobachgs left a comment

Choose a reason for hiding this comment

mvidner commented Jul 4, 2024 •

edited

Loading

coveralls commented Jul 29, 2024 •

edited

Loading