mirror of
https://github.com/windmill-labs/windmill.git
synced 2026-10-17 08:03:06 +00:00
GuilhemandClaude Opus 5.5
35ec16d127
* fix: delete a flow chat turn with its flow run, not its step jobs Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> * fix: give flow-written answers and overlapping turns their own flow run Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> --------- Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
419 lines
16 KiB
Rust
419 lines
16 KiB
Rust
use std::collections::HashMap;
|
|
|
|
use chrono::{DateTime, Utc};
|
|
use serde::{Deserialize, Serialize};
|
|
use serde_json::value::RawValue;
|
|
use sqlx::{self, FromRow};
|
|
use uuid::Uuid;
|
|
use windmill_types::s3::S3Object;
|
|
|
|
use crate::db::DB;
|
|
use crate::error::Result;
|
|
use crate::utils::truncate_with_ellipsis;
|
|
|
|
/// Changing it detaches every memory stored under a string memory id.
|
|
const MEMORY_ID_NAMESPACE: Uuid = Uuid::from_u128(0x6f1c2d4e_8a3b_5c7d_9e0f_1a2b3c4d5e6f);
|
|
|
|
/// Memory is stored and carried in `flow_status.memory_id` as a uuid, which names the same memory
|
|
/// wherever it is passed, as a chat conversation id must. Any other string names a memory through a
|
|
/// name-based (v5) uuid scoped to its workspace and flow, so the same key in two flows or two
|
|
/// workspaces names two memories, and chat conversation ids stay unique across workspaces.
|
|
pub fn memory_key(workspace_id: &str, flow_path: &str, memory_id: &str) -> Uuid {
|
|
let memory_id = memory_id.trim();
|
|
Uuid::parse_str(memory_id).unwrap_or_else(|_| {
|
|
use sha1::{Digest, Sha1};
|
|
let mut hasher = Sha1::new();
|
|
hasher.update(MEMORY_ID_NAMESPACE.as_bytes());
|
|
for part in [workspace_id, flow_path, memory_id] {
|
|
hasher.update(part.as_bytes());
|
|
hasher.update([0u8]);
|
|
}
|
|
let mut bytes = [0u8; 16];
|
|
bytes.copy_from_slice(&hasher.finalize()[..16]);
|
|
uuid::Builder::from_sha1_bytes(bytes).into_uuid()
|
|
})
|
|
}
|
|
|
|
#[derive(Debug, Clone, Copy, Serialize, Deserialize, PartialEq, Eq, sqlx::Type)]
|
|
#[sqlx(type_name = "MESSAGE_TYPE", rename_all = "lowercase")]
|
|
#[serde(rename_all = "lowercase")]
|
|
pub enum MessageType {
|
|
User,
|
|
Assistant,
|
|
System,
|
|
Tool,
|
|
}
|
|
|
|
#[derive(Serialize, FromRow, Debug)]
|
|
pub struct FlowConversation {
|
|
pub id: Uuid,
|
|
pub workspace_id: String,
|
|
pub flow_path: String,
|
|
pub title: Option<String>,
|
|
pub created_at: DateTime<Utc>,
|
|
pub updated_at: DateTime<Utc>,
|
|
pub created_by: String,
|
|
/// Started from the flow editor's test panel rather than a deployed run.
|
|
pub is_test: bool,
|
|
}
|
|
|
|
/// `is_test` is written on insert. An existing conversation of the other kind refuses the
|
|
/// turn, so preview and deployed runs never share one.
|
|
pub async fn get_or_create_conversation_with_id(
|
|
tx: &mut sqlx::Transaction<'_, sqlx::Postgres>,
|
|
w_id: &str,
|
|
flow_path: &str,
|
|
username: &str,
|
|
title: &str,
|
|
conversation_id: Uuid,
|
|
is_test: bool,
|
|
) -> Result<FlowConversation> {
|
|
if let Some(existing) = lock_conversation(tx, w_id, conversation_id).await? {
|
|
let existing = same_kind(existing, is_test)?;
|
|
refuse_running_turn(tx, conversation_id).await?;
|
|
return Ok(existing);
|
|
}
|
|
|
|
// Truncate title to 25 characters max
|
|
let title = truncate_with_ellipsis(title, 25);
|
|
|
|
// Every turn released by the same collector's commit finds no row: the first insert
|
|
// wins, the others wait on it, do nothing, and read the row it created.
|
|
let created = sqlx::query_as!(
|
|
FlowConversation,
|
|
"INSERT INTO flow_conversation (id, workspace_id, flow_path, created_by, title, is_test)
|
|
VALUES ($1, $2, $3, $4, $5, $6)
|
|
ON CONFLICT (id) DO NOTHING
|
|
RETURNING id, workspace_id, flow_path, title, created_at, updated_at, created_by, is_test",
|
|
conversation_id,
|
|
w_id,
|
|
flow_path,
|
|
username,
|
|
title,
|
|
is_test
|
|
)
|
|
.fetch_optional(&mut **tx)
|
|
.await?;
|
|
if let Some(conversation) = created {
|
|
return Ok(conversation);
|
|
}
|
|
|
|
// The concurrent first turn that won the insert may have been of the other kind.
|
|
let existing = lock_conversation(tx, w_id, conversation_id)
|
|
.await?
|
|
.ok_or_else(|| {
|
|
crate::error::Error::BadRequest(format!(
|
|
"conversation {conversation_id} belongs to another workspace"
|
|
))
|
|
})?;
|
|
let existing = same_kind(existing, is_test)?;
|
|
refuse_running_turn(tx, conversation_id).await?;
|
|
Ok(existing)
|
|
}
|
|
|
|
/// The turn a conversation is still answering: its newest user message, while the flow run
|
|
/// that message started is still queued or running.
|
|
#[derive(Serialize, Debug, Clone, Copy)]
|
|
pub struct RunningTurn {
|
|
pub job_id: Uuid,
|
|
/// `created_seq` of the user message that started the turn.
|
|
pub user_seq: i64,
|
|
}
|
|
|
|
/// One running turn per conversation holds its agent memory; a second run would write the
|
|
/// same memory concurrently. Checked under the conversation's row lock, so two runs sent at
|
|
/// once cannot both pass: the second waits, then sees the first's message and queued job.
|
|
async fn refuse_running_turn(
|
|
tx: &mut sqlx::Transaction<'_, sqlx::Postgres>,
|
|
conversation_id: Uuid,
|
|
) -> Result<()> {
|
|
let Some(turn) = running_turns(&mut **tx, &[conversation_id])
|
|
.await?
|
|
.remove(&conversation_id)
|
|
else {
|
|
return Ok(());
|
|
};
|
|
// A JSON body, so a chat client can follow the running turn instead of failing.
|
|
Err(crate::error::Error::Generic(
|
|
axum::http::StatusCode::CONFLICT,
|
|
serde_json::json!({
|
|
"error": "this conversation is still answering a message; wait for it to finish or stop it before sending another",
|
|
"running_turn": turn,
|
|
})
|
|
.to_string(),
|
|
))
|
|
}
|
|
|
|
/// The running turn of each of `conversation_ids` that has one.
|
|
///
|
|
/// It answers for whatever ids it is given and checks no permission of its own, so the
|
|
/// executor must be one the caller is entitled to read those conversations through: a
|
|
/// `user_db` transaction under RLS, or a transaction holding ids the caller has already
|
|
/// authorized. Handed a raw pool and ids from a request, it would report other users' jobs.
|
|
pub async fn running_turns<'e, E: sqlx::PgExecutor<'e>>(
|
|
executor: E,
|
|
conversation_ids: &[Uuid],
|
|
) -> Result<std::collections::HashMap<Uuid, RunningTurn>> {
|
|
let rows = sqlx::query!(
|
|
r#"SELECT c.id AS "id!", u.job_id AS "job_id!", u.created_seq AS "user_seq!"
|
|
FROM unnest($1::uuid[]) AS c(id)
|
|
CROSS JOIN LATERAL (
|
|
SELECT job_id, created_seq
|
|
FROM flow_conversation_message
|
|
WHERE conversation_id = c.id AND message_type = 'user'
|
|
ORDER BY created_seq DESC
|
|
LIMIT 1
|
|
) u
|
|
WHERE u.job_id IS NOT NULL
|
|
AND EXISTS (SELECT 1 FROM v2_job_queue q WHERE q.id = u.job_id)"#,
|
|
conversation_ids
|
|
)
|
|
.fetch_all(executor)
|
|
.await?;
|
|
Ok(rows
|
|
.into_iter()
|
|
.map(|r| (r.id, RunningTurn { job_id: r.job_id, user_seq: r.user_seq }))
|
|
.collect())
|
|
}
|
|
|
|
/// `memory_id` is the caller's to choose, so a preview run could name a deployed
|
|
/// conversation and the reverse. A conversation's kind is fixed at creation and nothing
|
|
/// would show the mixing afterwards, so the turn is refused before it starts.
|
|
fn same_kind(existing: FlowConversation, is_test: bool) -> Result<FlowConversation> {
|
|
if existing.is_test == is_test {
|
|
return Ok(existing);
|
|
}
|
|
Err(crate::error::Error::BadRequest(if existing.is_test {
|
|
"this conversation was started from the flow editor's test panel; start a new conversation to run the deployed flow".to_string()
|
|
} else {
|
|
"this conversation belongs to the deployed flow; start a new conversation to test from the flow editor".to_string()
|
|
}))
|
|
}
|
|
|
|
/// Locked, so a turn orders against retention collecting the conversation
|
|
/// (windmill_common::jobs::delete_jobs): either the turn goes first and the collector then
|
|
/// sees its message, or it waits and finds the row gone and creates it again. Unlocked, the
|
|
/// message insert would wait on the parent row's lock instead and then fail its FK check.
|
|
async fn lock_conversation(
|
|
tx: &mut sqlx::Transaction<'_, sqlx::Postgres>,
|
|
w_id: &str,
|
|
conversation_id: Uuid,
|
|
) -> Result<Option<FlowConversation>> {
|
|
Ok(sqlx::query_as!(
|
|
FlowConversation,
|
|
"SELECT id, workspace_id, flow_path, title, created_at, updated_at, created_by, is_test
|
|
FROM flow_conversation
|
|
WHERE id = $1 AND workspace_id = $2
|
|
FOR UPDATE",
|
|
conversation_id,
|
|
w_id
|
|
)
|
|
.fetch_optional(&mut **tx)
|
|
.await?)
|
|
}
|
|
|
|
/// What a row carries beyond its text. A chat is rebuilt from its rows alone, without
|
|
/// reading jobs, so every tool row carries the model's call and what the model got back: a
|
|
/// Windmill tool's job holds the args its input transforms produced rather than the model's,
|
|
/// and an MCP tool's call sits among every call of the turn in the agent's job. That job's
|
|
/// `reasoning` is one string for the whole turn, where the rows keep it per iteration.
|
|
#[derive(Debug, Clone, Default)]
|
|
pub struct MessageExtras {
|
|
pub tool_arguments: Option<String>,
|
|
pub tool_result: Option<String>,
|
|
pub reasoning: Option<String>,
|
|
/// The files a user message carried; see `message_attachments`.
|
|
pub attachments: Vec<MessageAttachment>,
|
|
}
|
|
|
|
/// The most files a user message keeps references to; the rest are dropped.
|
|
pub const MAX_MESSAGE_ATTACHMENTS: usize = 20;
|
|
|
|
/// A file a user message carried, as the object-storage reference its run received.
|
|
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
|
|
pub struct MessageAttachment {
|
|
/// The flow input that held it.
|
|
pub input: String,
|
|
pub s3: String,
|
|
#[serde(skip_serializing_if = "Option::is_none")]
|
|
pub storage: Option<String>,
|
|
#[serde(skip_serializing_if = "Option::is_none")]
|
|
pub filename: Option<String>,
|
|
}
|
|
|
|
/// The files a run's args carry for its user message: every top-level input other than
|
|
/// `user_message` whose value is an object-storage reference or a list of them, in input
|
|
/// name order, capped at `MAX_MESSAGE_ATTACHMENTS`. Only the reference is kept: `presigned`
|
|
/// grants access to the file, and any other value may be the file's bytes.
|
|
pub fn message_attachments(args: &HashMap<String, Box<RawValue>>) -> Vec<MessageAttachment> {
|
|
let mut inputs: Vec<_> = args
|
|
.iter()
|
|
.filter(|(name, _)| name.as_str() != "user_message")
|
|
.collect();
|
|
inputs.sort_by(|a, b| a.0.cmp(b.0));
|
|
inputs
|
|
.into_iter()
|
|
.flat_map(|(name, value)| {
|
|
serde_json::from_str::<S3Object>(value.get())
|
|
.map(|object| vec![object])
|
|
.or_else(|_| serde_json::from_str::<Vec<S3Object>>(value.get()))
|
|
.unwrap_or_default()
|
|
.into_iter()
|
|
.filter(|object| !object.s3.is_empty())
|
|
.map(move |object| MessageAttachment {
|
|
input: name.clone(),
|
|
s3: object.s3,
|
|
storage: object.storage,
|
|
filename: object.filename,
|
|
})
|
|
})
|
|
.take(MAX_MESSAGE_ATTACHMENTS)
|
|
.collect()
|
|
}
|
|
|
|
/// Add a message to a conversation using an existing transaction
|
|
/// If the conversation doesn't exist, logs a warning and returns Ok (no error thrown)
|
|
/// This allows memory_id to be used for agent memory without requiring a conversation
|
|
pub async fn add_message_to_conversation_tx(
|
|
tx: &mut sqlx::Transaction<'_, sqlx::Postgres>,
|
|
conversation_id: Uuid,
|
|
job_id: Option<Uuid>,
|
|
// The flow run the user message started; retention deletes the turn's rows by it.
|
|
turn_job_id: Uuid,
|
|
content: &str,
|
|
message_type: MessageType,
|
|
step_name: Option<&str>,
|
|
success: bool,
|
|
extras: Option<&MessageExtras>,
|
|
) -> Result<()> {
|
|
// Check if conversation exists first
|
|
let conversation_exists = sqlx::query!(
|
|
"SELECT EXISTS(SELECT 1 FROM flow_conversation WHERE id = $1) as \"exists!\"",
|
|
conversation_id
|
|
)
|
|
.fetch_one(&mut **tx)
|
|
.await?
|
|
.exists;
|
|
|
|
if !conversation_exists {
|
|
tracing::warn!(
|
|
"Conversation {} does not exist. Skipping message insertion. This is expected when flows are called from apps (memory_id is used for agent memory only).",
|
|
conversation_id
|
|
);
|
|
return Ok(());
|
|
}
|
|
|
|
// Insert the message
|
|
sqlx::query!(
|
|
"INSERT INTO flow_conversation_message (conversation_id, message_type, content, job_id, turn_job_id, step_name, success, tool_arguments, tool_result, reasoning, attachments)
|
|
VALUES ($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11)",
|
|
conversation_id,
|
|
message_type as MessageType,
|
|
content,
|
|
job_id,
|
|
turn_job_id,
|
|
step_name,
|
|
success,
|
|
extras.and_then(|e| e.tool_arguments.as_deref()),
|
|
extras.and_then(|e| e.tool_result.as_deref()),
|
|
extras.and_then(|e| e.reasoning.as_deref()),
|
|
extras
|
|
.map(|e| &e.attachments)
|
|
.filter(|attachments| !attachments.is_empty())
|
|
.map(sqlx::types::Json) as Option<sqlx::types::Json<&Vec<MessageAttachment>>>
|
|
)
|
|
.execute(&mut **tx)
|
|
.await?;
|
|
|
|
// Update conversation updated_at timestamp
|
|
sqlx::query!(
|
|
"UPDATE flow_conversation SET updated_at = NOW() WHERE id = $1",
|
|
conversation_id
|
|
)
|
|
.execute(&mut **tx)
|
|
.await?;
|
|
|
|
Ok(())
|
|
}
|
|
|
|
/// Delete all memory for a conversation from the database
|
|
pub async fn delete_conversation_memory(
|
|
db: &DB,
|
|
workspace_id: &str,
|
|
conversation_id: Uuid,
|
|
) -> Result<()> {
|
|
sqlx::query!(
|
|
"DELETE FROM ai_agent_memory WHERE workspace_id = $1 AND conversation_id = $2",
|
|
workspace_id,
|
|
conversation_id
|
|
)
|
|
.execute(db)
|
|
.await?;
|
|
|
|
Ok(())
|
|
}
|
|
|
|
#[cfg(test)]
|
|
mod tests {
|
|
use super::*;
|
|
use serde_json::{json, value::to_raw_value};
|
|
|
|
fn args(values: serde_json::Value) -> HashMap<String, Box<RawValue>> {
|
|
values
|
|
.as_object()
|
|
.unwrap()
|
|
.iter()
|
|
.map(|(name, value)| (name.clone(), to_raw_value(value).unwrap()))
|
|
.collect()
|
|
}
|
|
|
|
#[test]
|
|
fn keeps_only_object_storage_references() {
|
|
let attachments = message_attachments(&args(json!({
|
|
"user_message": { "s3": "not/an/attachment.png" },
|
|
"avatar": { "s3": "u/a.png", "storage": "secondary", "presigned": "https://signed" },
|
|
"files": [
|
|
{ "s3": "u/b.pdf", "filename": "b.pdf" },
|
|
{ "s3": "" }
|
|
],
|
|
"photo": "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mNkYPhfDwAChwGA60e6kgAAAABJRU5ErkJggg==",
|
|
"count": 3
|
|
})));
|
|
assert_eq!(
|
|
serde_json::to_value(&attachments).unwrap(),
|
|
json!([
|
|
{ "input": "avatar", "s3": "u/a.png", "storage": "secondary" },
|
|
{ "input": "files", "s3": "u/b.pdf", "filename": "b.pdf" }
|
|
])
|
|
);
|
|
}
|
|
|
|
#[test]
|
|
fn caps_the_references_of_one_message() {
|
|
let files: Vec<_> = (0..25)
|
|
.map(|i| json!({ "s3": format!("u/{i}.png") }))
|
|
.collect();
|
|
let attachments = message_attachments(&args(json!({ "files": files })));
|
|
assert_eq!(attachments.len(), MAX_MESSAGE_ATTACHMENTS);
|
|
assert_eq!(attachments.last().unwrap().s3, "u/19.png");
|
|
}
|
|
|
|
/// A string names a memory only within its workspace and flow; a uuid is used as is.
|
|
#[test]
|
|
fn memory_key_scopes_strings_but_not_uuids() {
|
|
let key = memory_key("ws", "f/support/triage", " customer-1 ");
|
|
assert_eq!(key, memory_key("ws", "f/support/triage", "customer-1"));
|
|
assert_ne!(
|
|
key,
|
|
memory_key("other_ws", "f/support/triage", "customer-1")
|
|
);
|
|
assert_ne!(key, memory_key("ws", "f/sales/triage", "customer-1"));
|
|
let conversation = Uuid::from_u128(7).to_string();
|
|
assert_eq!(memory_key("ws", "f/a", &conversation), Uuid::from_u128(7));
|
|
assert_eq!(
|
|
memory_key("other_ws", "f/b", &conversation),
|
|
Uuid::from_u128(7)
|
|
);
|
|
}
|
|
}
|