[BUG] WITHOUTCOUNT crash - [MOD-6306] #4224
Description
Testing on v2.8.9, I experienced a crash:
7:79:M 16 Dec 2023 22:44:45.685 # <search> successRatio == 1 means heap is full
from OPT_Rewind() in optimizer_reader.c:
RS_LOG_ASSERT(successRatio < 1, "successRatio == 1 means heap is full");
I don't think this is a heap issue, and adding some logging, we can see below OPT_Rewind is called for the first time:
7:79:M 16 Dec 2023 22:44:45.685 # <module> qOpt->limit [24] qOpt->nf->limit [2]
7:79:M 16 Dec 2023 22:44:45.685 # <module> oi->offset [0]
7:79:M 16 Dec 2023 22:44:45.685 # <module> rc1 [0], break...
7:79:M 16 Dec 2023 22:44:45.685 # <module> heap_size(it->heap) [24] heap_count(it->heap) [2] it->offset [0] it->childEstimate [80]
7:79:M 16 Dec 2023 22:44:45.685 # <module> OPT_Rewind nf->offset [75]
7:79:M 16 Dec 2023 22:44:45.685 # <module> successRatio [1.000000] resultsCollectedSinceLast [2] optIt->lastLimitEstimate [2]
7:79:M 16 Dec 2023 22:44:45.685 # <search> successRatio == 1 means heap is full
lastLimitEstimate was initialized to 2 in NewOptimizerIterator() and when OPT_Rewind() is called for the first time, resultsCollectedSinceLast is the same as lastLimitEstimate, causing the assertion. This does not happen all the time, so I am still testing to see what conditions trigger this bug.