[P0] time_since_last_block never grows on chain halt when drift monitoring is enabled #34

Priority: P0 (bug) · Effort: S

Problem

When drift monitoring is enabled (--reference-node + --full-nodes), pkg/exporters/drift/drift.go:66 and :85 call m.UpdateLastBlockTime(chainID, time.Now()) on every successful poll, whether or not the height advanced.

If the chain halts, ev_metrics_time_since_last_block_seconds is reset to 0 every --polling-interval (default 10s), so a "no new block for N seconds" alert never fires. That alert is the main reason this metric exists.

Proposed fix

• Keep lastHeight uint64 on the drift exporter.
• Call UpdateLastBlockTime only when refHeight > lastHeight.
• Remove the per-full-node UpdateLastBlockTime call. A full node catching up doesn't mean the chain produced a block.

Pros / Cons

• ✅ Restores chain-halt detection; ~5 lines.
• ⚠️ None. The polling fallback from fix: ensure time_since_last_block accuracy with polling fallback #28 stays accurate to within one polling interval.

Acceptance criteria

• New pkg/exporters/drift/drift_test.go: an httptest JSON-RPC server returns a constant eth_blockNumber, and the test asserts time_since_last_block_seconds keeps growing.
• A height that goes backwards (reorg or endpoint switch) doesn't reset the gauge.