zero available memory may drop congestion control and deadlock event broker

[lidezhu]

What did you do?

When available memory == 0, the collector may skip sending congestion-control information for that changefeed.

On the event broker side, this leads to:

• available memory quota is not set, skip scan
• the broker stops scanning and stops sending new events
• the collector cannot make forward progress either
• the whole pipeline can stall in a deadlock-like state

What did you expect to see?

Even when available memory is 0, congestion-control should still be sent to the broker, so the broker can stay on the normal backpressure path instead of treating the quota as missing.

What did you see instead?

None

Versions of the cluster

Upstream TiDB cluster version (execute SELECT tidb_version(); in a MySQL client):

(paste TiDB cluster version here)

Upstream TiKV version (execute tikv-server --version):

(paste TiKV version here)

TiCDC version (execute cdc version):

(paste TiCDC version here)

[lidezhu]

It is fixed by #4030.