Transaction Isolation Levels

This skill teaches how to configure transaction isolation levels in Prisma to prevent race conditions and handle concurrent database access correctly.

---

• User mentions race conditions, concurrent updates, or dirty reads
• Working with financial transactions, inventory systems, or booking platforms
• Implementing operations that must maintain consistency under concurrent access
• User asks about Serializable, RepeatableRead, or ReadCommitted isolation
• Dealing with P2034 errors (transaction conflicts)

Key Isolation Levels:

1. Serializable - Strictest isolation, prevents all anomalies
2. RepeatableRead - Prevents dirty and non-repeatable reads
3. ReadCommitted - Prevents dirty reads only (default for most databases)
4. ReadUncommitted - No isolation (not recommended)

Common Concurrency Issues:

• Dirty Reads: Reading uncommitted changes from other transactions
• Non-Repeatable Reads: Same query returns different results within transaction
• Phantom Reads: New rows appear in repeated queries
• Lost Updates: Concurrent updates overwrite each other

When to Set Isolation:

• Financial operations (payments, transfers, refunds)
• Inventory management (stock reservations, order fulfillment)
• Booking systems (seat reservations, room bookings)
• Any operation requiring strict consistency

Phase 1: Identify Concurrency Risk

1. Analyze operation for concurrent access patterns
2. Determine what consistency guarantees are needed
3. Choose appropriate isolation level based on requirements

Phase 2: Configure Isolation Level

1. Set isolation level in transaction options
2. Implement proper error handling for conflicts
3. Add retry logic if appropriate

Phase 3: Handle Isolation Conflicts

1. Catch P2034 errors (transaction conflicts)
2. Retry with exponential backoff if appropriate
3. Return clear error messages to users

Level	Prevents	Use Cases	Trade-offs
Serializable	All anomalies	Financial transactions, critical inventory	Highest consistency, lowest concurrency, more P2034 errors
RepeatableRead	Dirty reads, non-repeatable reads	Reports, multi-step reads	Good balance, still allows phantom reads
ReadCommitted	Dirty reads only	Standard operations, high-concurrency	Highest concurrency, allows non-repeatable/phantom reads
ReadUncommitted	Nothing	Not recommended	Almost never appropriate

Serializable Example

await prisma.$transaction(
  async (tx) => {
    const account = await tx.account.findUnique({
      where: { id: accountId }
    });

    if (account.balance < amount) {
      throw new Error('Insufficient funds');
    }

    await tx.account.update({
      where: { id: accountId },
      data: { balance: { decrement: amount } }
    });

    await tx.transaction.create({
      data: {
        accountId,
        amount: -amount,
        type: 'WITHDRAWAL'
      }
    });
  },
  {
    isolationLevel: Prisma.TransactionIsolationLevel.Serializable
  }
);

RepeatableRead Example

await prisma.$transaction(
  async (tx) => {
    const user = await tx.user.findUnique({
      where: { id: userId },
      include: { orders: true }
    });

    const totalSpent = user.orders.reduce(
      (sum, order) => sum + order.amount,
      0
    );

    await tx.user.update({
      where: { id: userId },
      data: {
        tierLevel: calculateTier(totalSpent),
        lastCalculatedAt: new Date()
      }
    });
  },
  {
    isolationLevel: Prisma.TransactionIsolationLevel.RepeatableRead
  }
);

ReadCommitted Example

await prisma.$transaction(
  async (tx) => {
    await tx.log.create({
      data: {
        level: 'INFO',
        message: 'User logged in',
        userId
      }
    });

    await tx.user.update({
      where: { id: userId },
      data: { lastLoginAt: new Date() }
    });
  },
  {
    isolationLevel: Prisma.TransactionIsolationLevel.ReadCommitted
  }
);

Follow this decision tree:

Is this a financial operation (money, payments, credits)?

• YES → Use Serializable
• NO → Continue

Does the operation read data multiple times and require it to stay constant?

• YES → Use RepeatableRead
• NO → Continue

Is this a high-concurrency operation where conflicts are expensive?

• YES → Use ReadCommitted (or no explicit isolation)
• NO → Continue

Does the operation modify data based on a read within the transaction?

• YES → Use RepeatableRead minimum
• NO → Use ReadCommitted (or no explicit isolation)

Still unsure?

• Start with RepeatableRead for safety
• Monitor P2034 error rate
• Adjust based on actual concurrency patterns

P2034: Transaction Conflict

When using Serializable isolation, conflicts are common under concurrency:

async function transferWithRetry(
  fromId: string,
  toId: string,
  amount: number,
  maxRetries = 3
) {
  for (let attempt = 0; attempt < maxRetries; attempt++) {
    try {
      await prisma.$transaction(
        async (tx) => {
          const fromAccount = await tx.account.findUnique({
            where: { id: fromId }
          });

          if (fromAccount.balance < amount) {
            throw new Error('Insufficient funds');
          }

          await tx.account.update({
            where: { id: fromId },
            data: { balance: { decrement: amount } }
          });

          await tx.account.update({
            where: { id: toId },
            data: { balance: { increment: amount } }
          });
        },
        {
          isolationLevel: Prisma.TransactionIsolationLevel.Serializable,
          maxWait: 5000,
          timeout: 10000
        }
      );

      return { success: true };

    } catch (error) {
      if (error.code === 'P2034' && attempt < maxRetries - 1) {
        await new Promise(resolve =>
          setTimeout(resolve, Math.pow(2, attempt) * 100)
        );
        continue;
      }

      throw error;
    }
  }

  throw new Error('Transaction failed after max retries');
}

Key Elements:

• Retry loop with attempt counter
• Check for P2034 error code
• Exponential backoff between retries
• maxWait and timeout configuration
• Final error if all retries exhausted

Timeout Configuration

{
  isolationLevel: Prisma.TransactionIsolationLevel.Serializable,
  maxWait: 5000,
  timeout: 10000
}

• maxWait: Maximum time to wait for transaction to start (milliseconds)
• timeout: Maximum time for transaction to complete (milliseconds)

Higher isolation levels need higher timeouts to handle conflicts.

MUST:

• Use Serializable for financial operations
• Handle P2034 errors explicitly
• Set appropriate maxWait and timeout values
• Validate data before starting transaction
• Use atomic operations (increment/decrement) when possible

SHOULD:

• Implement retry logic with exponential backoff for Serializable
• Keep transactions as short as possible
• Read all data needed before making decisions
• Log isolation conflicts for monitoring
• Consider RepeatableRead before defaulting to Serializable

NEVER:

• Use ReadUncommitted in production
• Ignore P2034 errors
• Retry indefinitely without limit
• Mix isolation levels in same operation
• Assume isolation level is higher than default without setting it

After implementing isolation levels:

1. Concurrency Testing:

   • Simulate concurrent requests to same resource
   • Verify no lost updates or race conditions occur
   • Expected: Conflicts detected and handled gracefully

2. Performance Monitoring:

   • Monitor P2034 error rate
   • Track transaction retry attempts
   • If P2034 > 5%: Consider lowering isolation level or optimizing transaction duration

3. Error Handling:

   • Verify P2034 errors return user-friendly messages
   • Check retry logic executes correctly
   • Ensure transactions eventually succeed or fail definitively

---

References

For additional details and advanced scenarios, see:

• Database-Specific Defaults - PostgreSQL, MySQL, SQLite, MongoDB isolation behaviors
• Race Condition Patterns - Lost updates, double-booking, phantom reads
• Complete Examples - Banking transfers, inventory reservations, seat bookings