Will TiDB query statements trigger the resolveLock operation?

Translated
1

Note:
This topic has been translated from a Chinese forum by GPT and might contain errors.

Original topic: tidb query语句会触发resolveLock操作吗

| username: llllllll

Version 5.3.0
Will TiDB query statements trigger the resolveLock operation? Currently, we have found that resolveLock takes a long time in many slow queries.

2
| username: cassblanca | Original post link

Query statements themselves do not trigger the resolveLock operation. If you find that resolveLock is taking a long time in slow queries, it may be that some queries in the application are being blocked, leading to long lock wait times. You need to look for the cause elsewhere.

3
| username: h5n1 | Original post link
  • ResolveLock:{num_rpc:1, total_time:12.117495ms}: The time taken to resolve the lock after encountering a lock while reading data. This usually occurs in scenarios with read-write conflicts.
4
| username: h5n1 | Original post link

When reading data or prewriting keys in a transaction, if there is already a Lock on the key, ResolveLock is needed. Why does this happen? There are several scenarios:

  1. Transaction txn_1 completes prewriting key_1 and key_2 but then exits abnormally. When transaction txn_2 tries to read key_1 and key_2, it will find the Lock written by txn_1 during prewrite.
  2. Transaction txn_1 completes prewriting key_1 but fails on key_2 due to a conflict. txn_1 will terminate and asynchronously clean up the lock on key_1. If the asynchronous cleanup is not yet complete, txn_2 will encounter the Lock when reading key_1.
  3. Transaction txn_1 successfully commits the primary key and then asynchronously commits the secondary keys. If the asynchronous commit is not yet complete, txn_2 will encounter the Lock when reading the secondary keys.

So how do you ResolveLocks? When prewriting keys, a LockTTL is also attached. The ResolveLock process first checks the TTL of all Locks, notes the longest expire Time, and if it finds any keys with expired locks, it initiates resolveLock for those expired keys. If there are still keys with unresolved locks, it will back off and retry based on the longest expire time.

5
| username: llllllll | Original post link

Thank you for the reply. So, the resolve lock will also be performed during the query, right? Wouldn’t it be better to perform the resolve lock during the write? Currently, it is at the snapshot isolation level. Lock conflicts will not affect read performance.

6
| username: h5n1 | Original post link

Yes.

7
| username: system | Original post link

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.