perf(validator): schedule metagraph sync and metagraph_history snapshots

[Thykof]

Problem

metagraph_history receives a full snapshot (one row per available uid) on every low-frequency prompt cycle, because the write is coupled to the prompt cadence. At short cycle intervals that is a large volume of near-identical rows per day with no consumer that needs that granularity — downstream analytics read the table on an hourly basis.

Change

Give the metagraph work its own schedule, decoupled from prompt cadence (constants in neurons/validator.py; each cycle runs as its own process with its own timer):

• Prompt processes (low/high): refresh the chain metagraph — miner_uids, axon addresses, and miner identity rows — at most every METAGRAPH_SYNC_INTERVAL_MINUTES = 15, writing no snapshot. The refresh runs after forward_prompt, so the chain sync never delays a minute-aligned prompt slot.
• Scoring process: the only metagraph_history writer, at most every METAGRAPH_SNAPSHOT_INTERVAL_MINUTES = 30 (including the boot snapshot; prompt-process restarts append nothing).
• Miner identities always upsert: insert_new_miners now runs on every refresh (the on-conflict updated_at bump is kept — it makes a re-current identity the newest row for its uid). Since miner_uids and the identity rows travel through the same call, a newly queryable uid always has its miners row before its first responses are saved.
• Degenerate chain responses: an empty sync result is applied (the shared metagraph object was just emptied — stale uids must not outlive it, forward_prompt skips cleanly) but the timer is not stamped, so the next cycle retries instead of waiting a full interval.
• New pure helper metagraph_refresh_due in synth/utils/helpers.py with unit tests; the save parameter is renamed to save_snapshot (identities are no longer gated by it) and made required.

Verification

• black --check / flake8 clean; new test_metagraph_refresh_due covers the gate boundaries (never-refreshed, below/at/above interval).
• Reviewed across multiple passes for the failure modes of the scheduling change: new/re-registered miner identity attribution, prompt-slot timing, empty-metagraph recovery, boot behavior per cycle, and caller compatibility of the signature change (single call site).

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR decouples metagraph syncing and metagraph_history snapshotting from the prompt cadence by introducing a dedicated refresh scheduler and a helper gate function, reducing unnecessary snapshot volume while keeping miner identity/address data fresh for querying.

Changes:

• Added metagraph_refresh_due() helper (with unit tests) to gate refresh operations by interval.
• Updated validator cycle loop(s) to refresh the metagraph on a separate schedule, writing metagraph_history snapshots only from the scoring process.
• Adjusted miner identity upserts so miner records are refreshed independently of snapshot writes, and renamed the forward helper parameter to save_snapshot.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
tests/test_helpers.py	Adds unit coverage for the new metagraph_refresh_due() interval gate.
synth/validator/forward.py	Renames save to required save_snapshot, always upserts miner identities, and gates metagraph_history writes by caller schedule.
synth/utils/helpers.py	Introduces metagraph_refresh_due() helper for refresh scheduling decisions.
neurons/validator.py	Implements per-cycle metagraph refresh scheduling and ensures only the scoring cycle writes snapshots.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.