The lock failure is really a timeout, and should be reported as such ("Roxie too busy - please try again later" or similar).

Once I address why this query takes so long the timeout should be much less frequent, of course.

I get the "Roxie control:lock failed" exception when there are 2 sendRoxieControlAllNodes calls at the same time. (The exception is from the 2nd call.)

The exception message is from: 

if (!sendRoxieControlLock(sock, allOrNothing, wait))
throw MakeStringException(-1, "Roxie control:lock failed");

Not sure what exception message should be changed to.

I get the empty response when I call the WuQueryGetGraph with an inactive (or invalid) query ID. Should roxie send back an exception saying that the query is inactive or invalid?

The following code is used to retrieve the data from roxie:

VStringBuffer control("<control:querystats><Query id='%s'/></control:querystats>", queryId);
Owned<IPropertyTree> querystats = sendRoxieControlAllNodes(eps.item(0), control.str(), false, ROXIELOCKCONNECTIONTIMEOUT);

ROXIELOCKCONNECTIONTIMEOUT is 60 seconds.

Two issues: taking too long in Roxie, and ESP not reporting well when Roxie is too slow to respond.

Seems as though the issue is that it is taking a very long time (and presumably, timing out). Seems to take significantly longer in "lock" mode.