Files
windmill/backend/windmill-api-groups/src/granular_acls.rs
Ruben Fiszel 0dbd9c1231 perf: eliminate dual-connection DB pool contention across worker, queue, and api (#9798)
* perf: eliminate dual-connection DB pool contention across worker, queue, and api

Reuse the held transaction (or move pool reads before begin()) instead of
checking out a second pool connection while a tx is open, extending the
fix from #9789/#7861. Targets the per-worker pool (max 5) hot paths plus
several server-pool API handlers.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix: pass owned pool to get_email_from_permissioned_as in http trigger handler

The generified signature takes impl PgExecutor; the http trigger handler
passed &db where db is already &DB, yielding &&Pool which does not impl
PgExecutor (only surfaced under the full feature set in CI).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix: keep RLS-exposed reads on the non-RLS pool and isolate flow-eval reads in a savepoint

Addresses review of the dual-connection sweep:

- worker_flow: wrap the stop_after_all_iters_if reads in a SAVEPOINT. The
  caller swallows the error and keeps using tx, so a DB read failure must
  not leave the outer transaction aborted (it would fail the later commit).
  Matches the previous pool-read semantics.

- Revert reads that were moved onto an RLS (user_db) transaction back to the
  non-RLS pool, since RLS row-visibility/role context can change results:
  push_scheduled_job (email/tag/settings lookups; reachable with a user_db
  tx from api-schedule/api-flows), push_inner native-retry dedicated_worker
  routing (RLS isolation variants), resources.rs app-namespace folder
  auto-create (non-admins must not be blocked), and the script archive/delete
  UPDATEs. Non-RLS db.begin() reuse and move-before-begin are kept.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* test: failpoint proving the stop_after_all_iters_if savepoint isolates an aborted read

Adds a worker-crate failpoints feature and a data-driven hook: when the
stop_after_all_iters_if expr is the magic sentinel, the in-evaluation read runs
SELECT 1/0 to abort its (savepoint) transaction. The test asserts the flow still
completes (iteration marked failed) — which only holds if the savepoint keeps the
outer status-update transaction committable. Without the savepoint the abort would
poison the outer tx and the job would never complete.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-25 21:33:17 +00:00

575 lines
19 KiB
Rust

/*
* Author: Ruben Fiszel
* Copyright: Windmill Labs, Inc 2022
* This file and its contents are licensed under the AGPLv3 License.
* Please see the included NOTICE for copyright information and
* LICENSE-AGPL for a copy of the license.
*/
use axum::{
extract::{Extension, Path},
routing::{get, post},
Json, Router,
};
use windmill_api_auth::require_owner_of_path;
use windmill_audit::audit_oss::audit_log;
use windmill_audit::ActionKind;
use windmill_common::scripts::ScriptHash;
use windmill_common::DB;
use windmill_git_sync::{handle_deployment_metadata, DeployedObject};
use windmill_api_auth::ApiAuthed;
use serde::{Deserialize, Serialize};
use windmill_common::{
db::UserDB,
error::{Error, JsonResult, Result},
utils::{not_found_if_none, StripPath},
};
/// Map a granular-ACL kind segment to the audit-log action prefix used by the
/// per-kind CRUD endpoints (e.g. `flows.update`, `scripts.update`). The
/// resulting action is suffixed with `.grant_acl` / `.revoke_acl` so the
/// audit log keeps a per-resource record of every `/acls/*` mutation —
/// folder/group already log via their dedicated permission-history tables.
fn audit_action_prefix_for_acl_kind(kind: &str) -> Option<&'static str> {
match kind {
"script" => Some("scripts"),
"flow" => Some("flows"),
"app" => Some("apps"),
// Distinct prefix so dashboards aggregating on `action` can separate
// raw_app ACL mutations from regular app ones without parsing the
// `kind` parameters field. (The granular_acls SQL routes raw_app
// writes to the same `app` table; audit log identity is separate.)
"raw_app" => Some("raw_apps"),
"resource" => Some("resources"),
"variable" => Some("variables"),
"schedule" => Some("schedules"),
"http_trigger" | "websocket_trigger" | "kafka_trigger" | "nats_trigger"
| "postgres_trigger" | "mqtt_trigger" | "gcp_trigger" | "azure_trigger" | "sqs_trigger"
| "email_trigger" => Some("triggers"),
_ => None,
}
}
const KINDS: [&str; 20] = [
"script",
"group_",
"resource",
"schedule",
"variable",
"flow",
"folder",
"app",
"raw_app",
"http_trigger",
"websocket_trigger",
"kafka_trigger",
"nats_trigger",
"postgres_trigger",
"mqtt_trigger",
"gcp_trigger",
"azure_trigger",
"sqs_trigger",
"email_trigger",
"volume",
];
pub fn workspaced_service() -> Router {
Router::new()
.route("/get/{*path}", get(get_granular_acls))
.route("/add/{*path}", post(add_granular_acl))
.route("/remove/{*path}", post(remove_granular_acl))
}
#[derive(Serialize, Deserialize)]
pub struct GranularAcl {
pub owner: String,
pub write: Option<bool>,
}
async fn add_granular_acl(
authed: ApiAuthed,
Extension(db): Extension<DB>,
Extension(user_db): Extension<UserDB>,
Path((w_id, path)): Path<(String, StripPath)>,
Json(GranularAcl { owner, write }): Json<GranularAcl>,
) -> Result<String> {
let path = path.to_path();
let (kind, path) = path
.split_once('/')
.ok_or_else(|| Error::BadRequest("Invalid path or kind".to_string()))?;
if !KINDS.contains(&kind) {
return Err(Error::BadRequest("Invalid kind".to_string()));
}
let identifier = if kind == "group_" || kind == "folder" || kind == "volume" {
"name"
} else {
"path"
};
if !authed.is_admin {
if kind == "folder" {
crate::folders::require_is_owner(&authed, path)?;
} else if kind == "group_" {
crate::groups::require_is_owner(path, &authed.username, &authed.groups, &w_id, &db)
.await?;
} else if kind == "volume" {
let created_by = sqlx::query_scalar!(
"SELECT created_by FROM volume WHERE name = $1 AND workspace_id = $2",
path,
&w_id
)
.fetch_optional(&db)
.await?
.ok_or_else(|| Error::NotFound(format!("volume '{path}' not found")))?;
// created_by is stored with u/ prefix (from job.permissioned_as)
let owner_username = created_by.strip_prefix("u/").unwrap_or(&created_by);
if owner_username != authed.username {
return Err(Error::NotAuthorized(
"Only the volume owner or an admin can modify permissions".to_string(),
));
}
} else {
require_owner_of_path(&authed, path)?;
}
}
let mut tx = user_db.begin(&authed).await?;
if kind == "folder" {
if let Some(obj) = sqlx::query_scalar!(
"SELECT owners FROM folder WHERE name = $1 AND workspace_id = $2",
path,
w_id
)
.fetch_optional(&mut *tx)
.await?
{
if obj.contains(&owner) {
if write != Some(true) {
return Err(Error::BadRequest(
"Cannot remove write permission for folder owner".to_string(),
));
}
}
}
}
// v2 raw apps are stored in the `app` table (with `app_version.raw_app = true`
// distinguishing them from regular apps); the legacy `raw_app` table no longer
// backs the workspace export, so granting/revoking on `raw_app` must hit `app`
// for the change to be visible. Git-sync dispatch still uses
// DeployedObject::RawApp so the worker writes back `<path>.raw_app.json`.
let table = if kind == "raw_app" { "app" } else { kind };
// SAFETY: `kind` has been validated against the `KINDS` allowlist before reaching this function.
let obj_o = sqlx::query_scalar::<_, serde_json::Value>(&format!(
"UPDATE {table} SET extra_perms = jsonb_set(extra_perms, $1, to_jsonb($2), \
true) WHERE {identifier} = $3 AND workspace_id = $4 RETURNING extra_perms"
))
.bind(vec![owner.clone()])
.bind(write.unwrap_or(false))
.bind(path)
.bind(&w_id)
.fetch_optional(&mut *tx)
.await?;
let _ = not_found_if_none(obj_o, &kind, &path)?;
if kind == "folder" {
let change_type = if write.unwrap_or(false) {
"grant_read"
} else {
"grant_write"
};
crate::folders::log_folder_permission_change(
&mut *tx,
&w_id,
path,
&authed.username,
change_type,
Some(&owner),
)
.await?;
} else if kind == "group_" {
let change_type = if write.unwrap_or(false) {
"grant_admin"
} else {
"grant_member_only"
};
crate::groups::log_group_permission_change(
&mut *tx,
&w_id,
path,
&authed.username,
change_type,
Some(&owner),
)
.await?;
} else if let Some(prefix) = audit_action_prefix_for_acl_kind(kind) {
// Mirror the folder/group permission-history coverage for every other
// ACLable kind. Folder/group already wrote a dedicated history row
// above; everything else (script/flow/app/raw_app/resource/...) lands
// in the general audit_log table here.
let access = if write.unwrap_or(false) {
"write"
} else {
"read"
};
let action = format!("{}.grant_acl", prefix);
audit_log(
&mut *tx,
&authed,
action.as_str(),
ActionKind::Update,
&w_id,
Some(path),
Some(
[
("kind", kind),
("owner", owner.as_str()),
("access", access),
]
.into(),
),
)
.await?;
}
tx.commit().await?;
match kind {
"folder" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::Folder { path: format!("f/{}", path) },
Some(format!("Folder '{}' changed permissions", path)),
true,
None,
)
.await?
}
"app" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::App { path: path.to_string(), parent_path: None, version: 0 },
Some(format!("App '{}' changed permissions", path)),
true,
None,
)
.await?
}
"raw_app" => {
// RawApp deliberately uses its own DeployedObject variant: the
// git-sync worker reads `path_type` ("app" vs "raw_app") to decide
// whether to write `<path>.app.json` or `<path>.raw_app.json`.
// Collapsing this into `App` would dispatch raw_app perm changes
// against the wrong file in the repo.
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::RawApp { path: path.to_string(), parent_path: None, version: 0 },
Some(format!("Raw App '{}' changed permissions", path)),
true,
None,
)
.await?
}
"script" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::Script {
path: path.to_string(),
parent_path: None,
hash: ScriptHash(0),
},
Some(format!("Script '{}' changed permissions", path)),
true,
None,
)
.await?
}
"flow" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::Flow { path: path.to_string(), parent_path: None, version: 0 },
Some(format!("Flow '{}' changed permissions", path)),
true,
None,
)
.await?
}
_ => (),
}
Ok("Successfully modified granular acl".to_string())
}
async fn remove_granular_acl(
authed: ApiAuthed,
Extension(db): Extension<DB>,
Extension(user_db): Extension<UserDB>,
Path((w_id, path)): Path<(String, StripPath)>,
Json(GranularAcl { owner, .. }): Json<GranularAcl>,
) -> Result<String> {
let path = path.to_path();
let (kind, path) = path
.split_once('/')
.ok_or_else(|| Error::BadRequest("Invalid path or kind".to_string()))?;
if !KINDS.contains(&kind) {
return Err(Error::BadRequest("Invalid kind".to_string()));
}
if !authed.is_admin {
if kind == "folder" {
crate::folders::require_is_owner(&authed, path)?;
} else if kind == "group_" {
crate::groups::require_is_owner(path, &authed.username, &authed.groups, &w_id, &db)
.await?;
} else if kind == "volume" {
let created_by = sqlx::query_scalar!(
"SELECT created_by FROM volume WHERE name = $1 AND workspace_id = $2",
path,
&w_id
)
.fetch_optional(&db)
.await?
.ok_or_else(|| Error::NotFound(format!("volume '{path}' not found")))?;
// created_by is stored with u/ prefix (from job.permissioned_as)
let owner_username = created_by.strip_prefix("u/").unwrap_or(&created_by);
if owner_username != authed.username {
return Err(Error::NotAuthorized(
"Only the volume owner or an admin can modify permissions".to_string(),
));
}
} else {
require_owner_of_path(&authed, path)?;
}
}
let mut tx = user_db.begin(&authed).await?;
let identifier = if kind == "group_" || kind == "folder" || kind == "volume" {
"name"
} else {
"path"
};
if identifier == "path" {
require_owner_of_path(&authed, path)?;
}
// See add_granular_acl: kind="raw_app" must hit the `app` table because v2
// raw apps live there and the legacy `raw_app` table no longer backs the
// workspace export.
let table = if kind == "raw_app" { "app" } else { kind };
// SAFETY: `kind` has been validated against the `KINDS` allowlist before reaching this function.
// LIMIT 1: `script` shares (workspace_id, path) across versions, so `old` can
// return >1 row, which would break the scalar subquery in RETURNING.
let obj_o = sqlx::query_scalar::<_, bool>(&format!(
"WITH old AS (
SELECT extra_perms->$1 as old_write FROM {table}
WHERE {identifier} = $2 AND workspace_id = $3 AND extra_perms ? $1
)
UPDATE {table} SET extra_perms = extra_perms - $1
WHERE {identifier} = $2 AND workspace_id = $3 AND extra_perms ? $1
RETURNING (SELECT old_write FROM old LIMIT 1)::bool"
))
.bind(&owner)
.bind(path)
.bind(&w_id)
.fetch_optional(&mut *tx)
.await?;
// Only log if something was actually removed (obj_o is Some)
if let Some(write) = obj_o {
// Log permission changes for folders and groups
if kind == "folder" {
let change_type = if write { "revoke_write" } else { "revoke_read" };
crate::folders::log_folder_permission_change(
&mut *tx,
&w_id,
path,
&authed.username,
change_type,
Some(&owner),
)
.await?;
} else if kind == "group_" {
crate::groups::log_group_permission_change(
&mut *tx,
&w_id,
path,
&authed.username,
"revoke_admin",
Some(&owner),
)
.await?;
} else if let Some(prefix) = audit_action_prefix_for_acl_kind(kind) {
// Mirror the add path: standard audit_log row for every kind that
// doesn't have a dedicated permission-history table.
let access = if write { "write" } else { "read" };
let action = format!("{}.revoke_acl", prefix);
audit_log(
&mut *tx,
&authed,
action.as_str(),
ActionKind::Update,
&w_id,
Some(path),
Some(
[
("kind", kind),
("owner", owner.as_str()),
("access", access),
]
.into(),
),
)
.await?;
}
tx.commit().await?;
match kind {
"folder" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::Folder { path: format!("f/{}", path) },
Some(format!("Folder '{}' changed permissions", path)),
true,
None,
)
.await?
}
"app" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::App { path: path.to_string(), parent_path: None, version: 0 },
Some(format!("App '{}' changed permissions", path)),
true,
None,
)
.await?
}
"raw_app" => {
// See add_granular_acl: raw_app must use its own DeployedObject
// variant so git-sync writes `<path>.raw_app.json`.
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::RawApp {
path: path.to_string(),
parent_path: None,
version: 0,
},
Some(format!("Raw App '{}' changed permissions", path)),
true,
None,
)
.await?
}
"script" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::Script {
path: path.to_string(),
parent_path: None,
hash: ScriptHash(0),
},
Some(format!("Script '{}' changed permissions", path)),
true,
None,
)
.await?
}
"flow" => {
handle_deployment_metadata(
&authed.email,
&authed.username,
&db,
&w_id,
DeployedObject::Flow { path: path.to_string(), parent_path: None, version: 0 },
Some(format!("Flow '{}' changed permissions", path)),
true,
None,
)
.await?
}
_ => (),
}
}
Ok("Successfully removed granular acl".to_string())
}
async fn get_granular_acls(
authed: ApiAuthed,
Extension(user_db): Extension<UserDB>,
Path((w_id, path)): Path<(String, StripPath)>,
) -> JsonResult<serde_json::Value> {
let path = path.to_path();
let (kind, path) = path
.split_once('/')
.ok_or_else(|| Error::BadRequest("Invalid path or kind".to_string()))?;
if !KINDS.contains(&kind) {
return Err(Error::BadRequest("Invalid kind".to_string()));
}
let mut tx = user_db.begin(&authed).await?;
let identifier = if kind == "group_" || kind == "folder" || kind == "volume" {
"name"
} else {
"path"
};
// See add_granular_acl: raw_app rows live in the `app` table now, so the
// read path must also target `app` — otherwise GET would return stale or
// 404 state while POST /acls/add and /acls/remove write to `app`.
let table = if kind == "raw_app" { "app" } else { kind };
// SAFETY: `kind` has been validated against the `KINDS` allowlist before reaching this function.
let obj_o = sqlx::query_scalar::<_, serde_json::Value>(&format!(
"SELECT extra_perms from {table} WHERE {identifier} = $1 AND workspace_id = $2"
))
.bind(path)
.bind(w_id)
.fetch_optional(&mut *tx)
.await?;
let obj = not_found_if_none(obj_o, &kind, &path)?;
tx.commit().await?;
Ok(Json(obj))
}