Concurrency fix plan

Deadlocks and Lock-Wait Timeouts

We map which transactions fight over which rows and re-sequence the access pattern so contention disappears instead of being retried.

Discuss your workload

Diagnosis context

Deadlocks and lock-wait timeouts are ordering problems wearing an error costume. Retry loops hide them for months, then a traffic spike exposes them. We instrument the exact lock graph, identify the transaction that arrives last and waits longest, and re-sequence access so both sides get what they need without the fight.

What the fix plan covers

Every intervention is reversible and measured.

01

Lock graph analysis

Deadlog and innodb lock instrumentation decoded into a map of who waits on whom.

02

Transaction re-sequencing

Consistent lock ordering and shorter critical sections — recommendations your team can apply in one commit.

03

Isolation review

Check whether the current isolation level is protecting you or manufacturing contention.

04

Retry hygiene

Where retries are legitimate, bound and instrument them; where they hide bugs, expose them.

Execution path

From live diagnosis to a measured, reversible fix.

  1. 01

    Capture deadlock logs and lock-wait instrumentation

  2. 02

    Map the contended resources and access order

  3. 03

    Re-sequence transactions and verify on staging

  4. 04

    Roll out and watch lock metrics fall

  5. 05

    Document the new access contract for your team

Measured outcome

Concurrency that scales with traffic instead of collapsing under it — with the contention map your team can reason about.

Common Questions

What engineering teams ask before working with us.

Our MySQL sits at 100% CPU right now. How fast can you help?

Emergency engagements start with a live diagnostic session — processlist sampling, digest analysis, and parameter review. Stabilization typically lands within one engagement day, with the root cause documented and a monitoring hook to catch recurrence.

Can a large ALTER really run without downtime?

Yes, when the strategy matches the change. Instant DDL covers many column operations in MySQL 8; in-place operations fit others; and gh-ost or pt-online-schema-change handle the rest with throttling tied to replica lag. We classify every change before anyone runs it, then rehearse on a production-size replica.

We keep getting deadlocks on the same tables. Is this a code problem?

It is an ordering problem, and the fix usually spans both sides. We map the lock graph and specify the transaction re-sequencing your team applies — typically a small, single-commit change. Isolation levels and index design, which are database-side, do most of the heavy lifting.

Why is our JSON-heavy API so slow?

Because queries filtering or sorting on JSON paths scan without an index. The fix keeps your flexible schema: virtual generated columns over the hot paths, indexes on those columns, and predicate rewrites at the database layer. Your application continues reading the same JSON.

Work with Burj

Bring your worst database incident to Burj.

Tell us what is slow — CPU, queries, migrations, or cost — and we will scope the smallest safe fix.

Plan an engagement