[Enhancement]: Extreme poll profile — meter/CT + PV live only

[Pho3niX90]

Summary

Proposal for a new extreme (minimal) poll profile aimed at tight control loops — e.g. export / backflow limiting against grid voltage rise (#451).

This is not a request to lower the global poll_interval_fast minimum to 1s for everyone. Solis protocol timing and our grouped Modbus reads make that unsafe on a full sensor set.

Why not just lower the global min?

Per RS485_MODBUS ESINV-33000ID Hybrid Inverter V3.1 §3.2:

More than 300ms communications frame interval is required. Recommended max data frame 100 bytes (50 registers).

Each sensor group is one Modbus frame. A typical hybrid FAST pass is ~11–12 groups → roughly 3.3–3.6s of spacing alone, plus transfer/dongle latency, plus normal/slow polls and writes on the same link.

Disabling HA entities does not stop group reads. essential_only already helps; this proposal goes further for control-loop users.

Proposed behaviour

New option (name TBD): extreme_only (or a poll profile enum: full / essential / extreme).

When enabled, only register these groups:

Group	Role	Entities (live)
33049	Solar	PV Voltage 1–4, PV Current 1–4, Total PV Power
33126	Meter / CT	Meter Voltage, Meter Current, Meter Active Power

Also keep ONCE identity groups (33000, 35000) so the HA device/serial setup still works.

Out of this profile: inverter A/B/C phase voltage (33070), battery, energy totals (as dedicated groups), settings entities (number/switch/select/time). Writes remain via existing services (solis_write_holding_register, dispatch, etc.).

Why meter, not phase voltage? For DNO / house voltage rise, CT-clamp meter voltage is the signal that matters more than inverter AC phase voltage.

Note: 33126 also includes Meter Total Active Energy in the same Modbus block — those registers are still read (can't skip mid-block without splitting the group).

Poll interval floor (proposal)

• Default / full / essential: keep poll_interval_fast min = 10s
• Extreme only: allow down to 2s (2 × ~300ms + transfer + occasional writes; not 1s)

Open questions for maintainers / community

1. Is extreme_only boolean better, or a single poll_profile select (full / essential / extreme)?
2. Is 2s the right floor, or prefer 3s / keep 10s and only shrink the group set?
3. Should ONCE identity groups stay in extreme mode?
4. String inverter equivalent later?
5. Dual-meter block (33300) in extreme, or never?
6. Anything else that must be in the minimal set for real-world export/voltage automations (e.g. household load, battery power)?

Workaround today (#451)

Until/unless this ships:

• Use essential_only + disable unused features to shrink bus load
• Drive the control loop with services: solis_read_register / solis_write_holding_register (and solis_dispatch where firmware supports it)

Related

• [Enhancement]: Update time #451 — request to lower min update time to 1s

[Pho3niX90]

@Nino6689 thoughts?

@cokeveryday @Gorwanvfr would this fit your use cases?

[Nino6689]

@Pho3niX90 I like this a lot — it's the right answer to #451. A tight control loop doesn't need more polling, it needs less surface polled faster, and profiles express that better than letting everyone turn the global floor down. Taking your open questions in order:

1. Boolean vs poll_profile select — select, definitely. full / essential / extreme as a single enum is clearer than two interacting booleans (essential_only + extreme_only would raise the question of what both-true means). Since essential_only is mine: happy for the enum to subsume it. Migration is cheap — in async_migrate_entry, map essential_only: true → poll_profile: "essential", absent/false → "full", and keep is_essential_only() as a thin wrapper meaning profile != "full" so nothing downstream churns. I'd also rather have the enum now than add a fourth mode later as a breaking change.

2. 2s floor — right, with one refinement. Two live frames is ~600ms of mandatory spacing plus transfer, so 2s leaves honest headroom for service writes on the same link. I wouldn't go to 3s by default, but the real variable isn't the frame budget — it's the dongle. Wired RS485/Elfin users will be fine at 2s; the WiFi sticks add latency that eats the margin. Rather than gating on that, I'd ship 2s and let the existing failed-read/backoff handling be the judge — a note in the option description ("2s assumes a wired or low-latency connection") covers it.

3. ONCE identity groups — keep them. They're a one-time cost at setup, and model/serial detection gates feature behaviour elsewhere; an extreme profile that produces a half-identified device in the registry would generate more issues than it saves frames.

4. String inverter equivalent — worth doing, but as a follow-up once the hybrid shape settles. Same pattern, different group numbers; no reason to hold this on it.

5. 33300 dual-meter — not until a dual-meter user asks with a concrete control use case. Every group added to extreme erodes the reason extreme exists.

6. What else the minimal set needs — this is the one place I'd push. For pure export-vs-voltage limiting, meter active power alone closes the loop and your two-group set is correct. But a lot of real-world export automations are battery-aware (mine included — surplus vs. SOC vs. dispatch decisions), and everything they need lives in one additional frame: group 33132 carries Battery SOC (33139), Household Load (33147), Backup Load (33148), Battery Power (33149/33150) and AC Grid Port Power (33151/33152). I'd make it opt-in rather than default — extreme plus an "include battery/load group" toggle, or a separate extreme_battery profile. Three frames is ~900ms of spacing, still comfortably inside a 2s floor. That single option is the difference between extreme mode serving only the #451 case and serving most battery-owning control-loop users.

One non-question note: worth stating in the docs that in extreme mode the excluded entities aren't "unavailable", they're simply not registered (same as essential_only today) — that framing preempts the "my sensors disappeared" reports, which after this week we've all had enough of.

Happy to pick up the implementation if useful. The profile enum plus migration sits right on top of the essential_only plumbing from #418, and I can extend the #448 construction matrix to build all three profiles against every inverter config so profile regressions fail in CI too.

[Nino6689]

One regulatory note that I think belongs in this decision, mainly because it bounds what extreme mode should claim.

Home Assistant can't be a compliant export limitation scheme at any poll rate. G100 requires an ELS to hold the agreed export limit "irrespective of the failure of one or more of its Components or the failure of any communications between the ELS's Components and Devices." A loop running in HA over Modbus — especially across a WiFi dongle — can't meet that by construction, since comms failure is precisely the failure mode it would have to survive. So if someone lands here wanting a faster interval for G100 compliance, the honest answer is that no interval fixes it; that needs type-tested hardware.

Which is another argument for keeping the global floor at 10s and putting the speed behind an opt-in profile. I'd suggest one line in the docs when this ships: extreme mode is for optimisation, not protection or compliance.

@Gorwanvfr — two things that may help you more than a faster poll:

Your inverter almost certainly isn't tripping at 253V. G98/G99 over-voltage protection stage 1 is typically around 262.2V with a 1s trip (exact figures depend on the grid-code profile your unit is set to). That protection exists to catch dangerous conditions, not to enforce supply-quality limits. 253V is the statutory limit on your DNO, not your trip threshold — so it's worth confirming what your inverter actually trips at, because you may have more headroom than you're currently leaving yourself.

And 253V is a duty they owe you. Under ESQCR 2002 Reg 27 the declared supply is 230V +10%/−6%, so if your voltage genuinely exceeds 253V that's a breach the DNO must investigate and remedy — usually a tap change on the local transformer, at no cost to you. Compliance is assessed on 10-minute means under BS EN 50160, which is why "it's still within limits" gets said so readily. Log 10-minute means rather than instantaneous peaks and raise it formally; instantaneous spikes are easy for them to wave away, sustained means are not. Throttling your own export may be solving their problem at your expense.

Tangential, but if your house sits near 250V routinely it's worth checking the Uc rating of any surge protection at your consumer unit. SPDs won't help with voltage rise at all — they're transient-only — but a standing voltage close to a MOV's Uc (typically 275V on UK single-phase) ages it faster than normal.

None of which changes the design here. Extreme mode is still worth having for exactly the reason you asked for it.

[Pho3niX90]

Perfect, if you would like to do it your are more than welcome since it touches one of your recent features.