Combo (BAG) market data silently returns NaN after the gateway's been up a while - legs still quote fine

[Marigold]

Running into something odd with futures calendar spreads and hoping someone's seen it.

I subscribe to a bunch of calendar spreads as BAG combos (-1 front + 1 back, exchange="SMART", native exchange on the legs). Works fine right after a gateway restart. But after the gateway's been running a
while, reqMktData on specific combos starts returning an empty ticker — bid/ask/last/close all NaN, ticker.ticks stays empty — and it never recovers on its own. No error callback, nothing.

The confusing part:

• The individual legs quote fine the whole time (same connection, reqMktData on each leg gives live bid/ask/size).
• Other combos on the same exchange quote fine simultaneously. Today I had NGQ26/NGV26 (NYMEX) and ZWZ6/ZWN7 (CBOT) both dead while NGN26/NGU26, NGU26/NGV26 and ZWN6/ZWU6 were all live. So it's
  per-conId, not per-exchange and not a farm outage.
• reqHistoricalData on the dead combos returns nothing either (though I used whatToShow=TRADES, so that one's not conclusive on its own).
• reqMarketDataType(2)/(3) (frozen/delayed) makes no difference — still empty.
• Market data farms all report OK (2104), no 354 / 10089, no "market data lines" error, nothing.

Only thing that fixes it is restarting the gateway.

ib.reqMarketDataType(1)
front = Contract(symbol="NG", secType="FUT", exchange="NYMEX", localSymbol="NGQ26")
back  = Contract(symbol="NG", secType="FUT", exchange="NYMEX", localSymbol="NGV26")
ib.qualifyContracts(front, back)
combo = Contract(symbol="NG", secType="BAG", exchange="SMART", currency="USD")
combo.comboLegs = [
    ComboLeg(conId=front.conId, ratio=1, action="SELL", exchange="NYMEX"),
    ComboLeg(conId=back.conId,  ratio=1, action="BUY",  exchange="NYMEX"),
]
t = ib.reqMktData(combo, "", False, False)
ib.sleep(5)
print(t.bid, t.ask, len(t.ticks))   # nan nan 0  (but the two legs quote fine)

I tried to reproduce it deliberately and couldn't, which is the frustrating bit:

• hammering one combo with 40x subscribe/cancel in a single connection → stays alive
• 25x connect → subscribe → cancel → immediate disconnect (no sleep, to race the cancel) → stays alive

So it doesn't seem tied to subscribe/cancel volume or to cancel-vs-disconnect races, at least not the way I can force it. The combos that died had only been subscribed ~2-3 times across separate processes
before going dark.

Environment: ib_async 2.1.0, Python 3.10.12, IB Gateway 10.45 (gnzsnz/ib-gateway docker image), Linux.

Is this a known gateway-side thing? Mostly wondering if anyone has a reliable repro or a workaround short of restarting, or knows whether it's worth reporting to IBKR directly since it smells server-side.

[mattsta]

Yeah, this is a years-long "ibkr gateway/data is broken weird" problem where sometimes it just refuses to quote prices on spreads even though all of the legs have prices (it also has bugs around sometimes when you add a valid contract it can refuse to qualify or recognize or transact it even though it really does exist until the gateway restarts and kicks out the bad data somehow!).

To avoid the "bag quotes end up dead" problem in my own client system, I created a fallback mechanism where:

• use the bag price if given
• if the bag price is bad (missing a bid and ask current with a Ticker update), manually calculate the bag bid/ask/sizes from all live leg quotes (which were previously registered to this mechanism via magic cross-attach-lookup-ownership logic)

This requires some careful management though because you must make sure you are always subscribe to market data for each leg in addition to the bag itself.

I have a Ticker wrapper called ITicker over there which does all the bag-quote-from-legs logic automatically:

• data structure for tracking legs and bags: https://github.com/mattsta/icli/blob/677da59e9d5077059f80de3e9e5a3a8c5e33a15e/icli/helpers.py#L706-L714

• math for updating bag price from legs: https://github.com/mattsta/icli/blob/677da59e9d5077059f80de3e9e5a3a8c5e33a15e/icli/helpers.py#L773-L809

• annoying meta-logic for attaching legs and bags to the meta-ITicker system on every quote subscribe or unsubscribe: https://github.com/mattsta/icli/blob/677da59e9d5077059f80de3e9e5a3a8c5e33a15e/icli/cli.py#L5211-L5269

• one benefit of also tracking legs per spread though: you can generate live tracking greeks on bags because the derived greeks for a spread are just additive greeks from each leg combined.

also in the same theme as all of this, I gave up on using "ibkr greeks" from their own data feeds because those values aren't updated quickly enough (especially around near-expiry where outsized moves and price decay start to matter every couple seconds) and instead of I update greeks for legs and whole bags on every tick with live time data as well (i don't think this is in the live repo update, it'll be out in the next feature snapshot maybe after I clear all the PDT handling from my client too).

[gnzsnz]

TWS suffers the same problem. Positions including single legs and combos can, maybe I should say will, show wrong/outdated "last" and PnL.

[Marigold]

Thanks a lot, it's good to confirm I'm not doing something completely stupid. I'll try the workaround with synthetic quote from legs based on your code (the library looks great).

[siongsheng]

Re-reading properly — you've already ruled out the farm outage (2104 OK, and it's per-conId not per-exchange), so ignore that angle. This matches what @mattsta said: the long-standing IBKR bug where the Gateway stops quoting specific combos while every leg still quotes fine. It's server-side and sticky until a restart — not something in your code, so at least you can stop chasing that.

Since the legs keep quoting the whole time, the synthetic-quote-from-legs route you mentioned is the pragmatic fix — derive the spread bid/ask from the leg quotes per your ratios instead of trusting the combo ticker. Sidesteps the dead combo without needing a restart.