Polygon sync enters feedback loop when milestones accumulate

[madumas]

Description

The Polygon sync service enters a feedback loop when milestones accumulate faster than they can be processed. This occurs after a node restart or during periods of slow execution.

Symptoms

• Sync cycle times increase progressively
• Node falls further behind despite continuous processing
• waypointsLen values grow each cycle
• age (time behind chain head) increases instead of decreasing

Logs demonstrating the problem

After a node restart, the following pattern emerges:

[02-03|14:54:39] [sync] downloading blocks using waypoints waypointsLen=94 ...
[02-03|15:00:44] head updated age=6m8s
[02-03|15:00:44] [sync] update fork choice done in=5m0s

[02-03|15:00:44] [sync] downloading blocks using waypoints waypointsLen=142 ...
[02-03|15:08:12] head updated age=7m34s
[02-03|15:08:12] [sync] update fork choice done in=6m30s

[02-03|15:08:12] [sync] downloading blocks using waypoints waypointsLen=159 ...

Analysis

Time	waypointsLen	Cycle duration	Age behind head
14:54:39	94	-	(restart)
15:00:44	142	5m00s	6m08s
15:08:12	159	6m30s	7m34s

Each cycle:

1. Processes all available waypoints (94 → 142 → 159)
2. Takes longer than the previous cycle (5m → 6m30s)
3. Falls further behind the chain head (6m → 7m34s)

The node accumulates milestones faster than it can process them, creating a feedback loop that prevents convergence.

Environment

• Chain: bor-mainnet
• Version: v3.3.7