Crash: thread 'main' panicked at 'Queue has spillover on Traversable::Lane ... sim/src/mechanics/queue.rs:91:17

[andrewguertin]

I'm getting a frequent crash at sim/src/mechanics/queue.rs#L91. It occurs after a few (game-time) hours of simulating huge_seattle with weekday_typical_traffic_from_psrc, if another crash doesn't happen first.

On Traversable::Lane(85385) at 07:56:35.6...
- CarID(38433 -- car) @ 84.9242m (length 4.8234m)
  Queued currently
- CarID(57086 -- car) @ 79.1008m (length 5.3613m)
  Queued currently
- CarID(4276 -- car) @ 72.7395m (length 6.1767m)
  Queued currently
- CarID(42418 -- car) @ 65.5628m (length 6.1847m)
  Queued currently
- CarID(140777 -- car) @ 44.0905m (length 4.8075m)
  Parking during 07:56:34.6 .. 07:56:49.6
- CarID(144640 -- car) @ 38.283m (length 5.1546m)
  Queued currently
- CarID(42154 -- car) @ 32.1284m (length 4.6616m)
  Queued currently
- CarID(134617 -- car) @ 15.6336m (length 5.2654m)
  Queued currently
- CarID(38281 -- car) @ 9.3682m (length 6.246m)
  Queued currently
- CarID(36155 -- car) @ 2.1222m (length 5.1077m)
  Queued currently

thread 'main' panicked at 'Queue has spillover on Traversable::Lane(85385) at 07:56:35.6 -- can't draw CarID(44507 -- car), bound is -3.9855m. Laggy head is Some(CarID(78903, Car))', sim/src/mechanics/queue.rs:91:17

Attached a few logs of the last minute of console output with backtrace.

queue.rs-91-1.txt
queue.rs-91-2.txt
queue.rs-91-3.txt

[dabreegster]

Thanks for the logs! The full weekday scenario on huge_seattle has loads of bugs -- this one is a known issue. (Getting a full day to complete without hitting permanent gridlock is only possible right now on the montlake map, which has lots of manual fixes to short roads / coupled intersections.)

I'm pretty sure this particular problem happens when there's a nice cycle of laggy heads (see https://github.com/dabreegster/abstreet/blob/master/docs/articles/trafficsim/article.md#exact-positions). I need to make an easy test case to repro this (using map_editor to create a synthetic map) and then think through what should happen here.

[dabreegster]

I happened to get an easy repro of this problem. When a car is half in a turn, half on the end of the last lane, they're a "laggy head. That influences the front position of the first car on a lane; the laggy head prevents the first in the queue from being at the very end of the lane. get_car_positions recurses to figure out exactly where the back of the laggy head is. I have to figure out how to break the recursion and decide where the back of a car is for this case

[dabreegster]

I think this particular issue is gone. https://github.com/dabreegster/abstreet/projects/2 the best way to track current work on getting simulations on large maps to work properly.