Potential OOM in PopFrontDDL Method of ddlPullerImpl

[wlwilliamx]

What did you do?

Problem

In the PopFrontDDL method of the ddlPullerImpl struct, the operation h.pendingDDLJobs = h.pendingDDLJobs[1:] is used to remove the first element from the pendingDDLJobs slice. However, this can lead to memory retention issues due to the behavior of Go slices:

• Slices in Go are references to the underlying array. When an element is removed by slicing, the reference to the element still exists in the underlying array, even though it is no longer part of the slice.
• As a result, the removed element (in this case, the first element of pendingDDLJobs) cannot be garbage collected, leading to unnecessary memory usage if the element is no longer needed and holds significant memory.

The problematic section of the code is as follows:

func (h *ddlPullerImpl) PopFrontDDL() (uint64, *timodel.Job) {
    h.mu.Lock()
    defer h.mu.Unlock()
    if len(h.pendingDDLJobs) == 0 {
        return atomic.LoadUint64(&h.resolvedTS), nil
    }
    job := h.pendingDDLJobs[0]
    h.pendingDDLJobs = h.pendingDDLJobs[1:] // Memory retention issue here
    return job.BinlogInfo.FinishedTS, job
}

Proposed Solution

To resolve this issue, the reference to the removed element should be explicitly set to nil before updating the slice. This will break the reference to the element, allowing the garbage collector to reclaim the memory.

Updated code:

func (h *ddlPullerImpl) PopFrontDDL() (uint64, *timodel.Job) {
    h.mu.Lock()
    defer h.mu.Unlock()
    if len(h.pendingDDLJobs) == 0 {
        return atomic.LoadUint64(&h.resolvedTS), nil
    }
    job := h.pendingDDLJobs[0]
    h.pendingDDLJobs[0] = nil // Clear reference to allow GC
    h.pendingDDLJobs = h.pendingDDLJobs[1:]
    return job.BinlogInfo.FinishedTS, job
}

This issue is not a memory leak but a memory retention problem due to Go’s slice mechanics. The fix should improve the memory efficiency of the system.

What did you expect to see?

No memory retention problem.

What did you see instead?

Memory retention problem in PopFrontDDL method.

Versions of the cluster

TiCDC version (execute cdc version):

(paste TiCDC version here)

[fubinzh]

/severity moderate

[wlwilliamx]

/affects 6.5 7.1 7.5 8.1 8.5