From 4b75c375dfd9b041d56d1a9d6bcabb0671e90bbf Mon Sep 17 00:00:00 2001
From: Pascal Jufer <pascal-jufer@bluewin.ch>
Date: Thu, 25 Apr 2024 14:19:17 +0200
Subject: [PATCH] Client-side restart of failed transactions on CockroachDB
 (#22240)

---
 .changeset/bright-mirrors-rush.md |  5 ++++
 api/src/utils/transaction.ts      | 47 +++++++++++++++++++++++++++++--
 2 files changed, 49 insertions(+), 3 deletions(-)
 create mode 100644 .changeset/bright-mirrors-rush.md

diff --git a/.changeset/bright-mirrors-rush.md b/.changeset/bright-mirrors-rush.md
new file mode 100644
index 0000000000..3b48da331c
--- /dev/null
+++ b/.changeset/bright-mirrors-rush.md
@@ -0,0 +1,5 @@
+---
+'@directus/api': patch
+---
+
+Implemented client-side restart of failed transactions for CockroachDB
diff --git a/api/src/utils/transaction.ts b/api/src/utils/transaction.ts
index 61e0783996..0b3c9930f9 100644
--- a/api/src/utils/transaction.ts
+++ b/api/src/utils/transaction.ts
@@ -1,4 +1,6 @@
-import type { Knex } from 'knex';
+import { type Knex } from 'knex';
+import { getDatabaseClient } from '../database/index.js';
+import { useLogger } from '../logger.js';
 
 /**
  * Execute the given handler within the current transaction or a newly created one
@@ -7,10 +9,49 @@ import type { Knex } from 'knex';
  * Can be used to ensure the handler is run within a transaction,
  * while preventing nested transactions.
  */
-export const transaction = <T = unknown>(knex: Knex, handler: (knex: Knex) => Promise<T>): Promise<T> => {
+export const transaction = async <T = unknown>(knex: Knex, handler: (knex: Knex) => Promise<T>): Promise<T> => {
 	if (knex.isTransaction) {
 		return handler(knex);
 	} else {
-		return knex.transaction((trx) => handler(trx));
+		try {
+			return await knex.transaction((trx) => handler(trx));
+		} catch (error: any) {
+			const client = getDatabaseClient(knex);
+
+			/**
+			 * This error code indicates that the transaction failed due to another
+			 * concurrent or recent transaction attempting to write to the same data.
+			 * This can usually be solved by restarting the transaction on client-side
+			 * after a short delay, so that it is executed against the latest state.
+			 *
+			 * @link https://www.cockroachlabs.com/docs/stable/transaction-retry-error-reference
+			 */
+			const COCKROACH_RETRY_ERROR_CODE = '40001';
+
+			if (client !== 'cockroachdb' || error?.code !== COCKROACH_RETRY_ERROR_CODE) throw error;
+
+			const MAX_ATTEMPTS = 3;
+			const BASE_DELAY = 100;
+
+			const logger = useLogger();
+
+			for (let attempt = 0; attempt < MAX_ATTEMPTS; ++attempt) {
+				const delay = 2 ** attempt * BASE_DELAY;
+
+				await new Promise((resolve) => setTimeout(resolve, delay));
+
+				logger.trace(`Restarting failed transaction (attempt ${attempt + 1}/${MAX_ATTEMPTS})`);
+
+				try {
+					return await knex.transaction((trx) => handler(trx));
+				} catch (error: any) {
+					if (error?.code !== COCKROACH_RETRY_ERROR_CODE) throw error;
+				}
+			}
+
+			/** Initial execution + additional attempts */
+			const attempts = 1 + MAX_ATTEMPTS;
+			throw new Error(`Transaction failed after ${attempts} attempts`, { cause: error });
+		}
 	}
 };