Symmetrical Simulation Freezes at Point of Reversal

[pjbollinger]

There is an issue with the sim command when using the --symmetrical option. It will simulate the history of trades like normal. However, once it reaches the point of reversing, it appears to get caught in a loop. I used mongoreplay to capture this log (an excerpt of the full log) and a snippet at time of reversal. There is a lot of data in there but you can see at the point of reversal there a couple of find orders in a row and then starts alternating find and getMore orders. After the reversal, it keeps repeating the same find and getMore orders though.

While debugging #750, I found that the program kept entering this logic of the code, which I think makes sense. However, the cursor does not appear to be moving correctly. I will be investigating this but wanted to share my findings so far.

[pjbollinger]

Here is some data of the queries around the code that identifies the original data is complete and it begins the reversal process. Notice how the $lt selector for the time freezes at the end.

Query (opts) at point of entering if (!trades.length) { if (so.symmetrical && !reversing) {...}}:

{ query:
  { selector: 'gdax.ETH-USD',
    time: { '$gt': 1511632608799 } },
  sort: { time: 1 },
  limit: 1000 }

Query (opts) at point of entering if (so.symmetrical && reversing) {...} the first time:

{ query: 
   { selector: 'gdax.ETH-USD',
     time: { '$lt': 1511632608799, '$gte': 1511373600000 } },
  sort: { time: -1 },
  limit: 1000 }

Query (opts) at point of entering if (so.symmetrical && reversing) {...} the second time:

{ query: 
   { selector: 'gdax.ETH-USD',
     time: { '$lt': 1511632981090, '$gte': 1511373600000 } },
  sort: { time: -1 },
  limit: 1000 }

Query (opts) at point of entering if (so.symmetrical && reversing) {...} the third time and forever after:

{ query: 
   { selector: 'gdax.ETH-USD',
     time: { '$lt': 1511632980913, '$gte': 1511373600000 } },
  sort: { time: -1 },
  limit: 1000 }

[pjbollinger]

I found the issue is caused by the cursor not properly adjusting from the reversal of time.

Here is a screenshot of a simulation HTML report after the fix #753:

When running, I recommend using --min_periods=0 to achieve near perfect symmetry.