-
Type: Bug
-
Resolution: Gone away
-
Priority: Minor - P4
-
None
-
Affects Version/s: None
-
Component/s: None
-
Storage Engines
edit: From the comments below:
When no config is specified test/format sets all table fields randomly including table_type which is one of FIX, ROW, and VAR. FIX (FLCS) isn't supported on the develop branch so this is a misuse of test/format.
To be determined if we add a warning when this happens, or trust devs to provide correct configs to test/format.
===================================
Compiling from the head of develop (current commit f35abcd) test/format is reliably reporting search mismatches such as the following:
table:T00009: WiredTiger trace ID: 10
snapshot-isolation 86552 search mismatch
expected {DUPLICATEV/LMNOPQRS\00\00\00\00\00\00nmlkji\00\00edcbazyi}
found {nUPLICATEV/LMNOPQRS\00\00\00\00\00\00nmlkji\00\00edcbazyi}
t: FAILED: snap_verify/337: 0
file:F00001: WiredTiger trace ID: 2
snapshot-isolation 195682 search mismatch
expected {D}
found {6}
t: FAILED: snap_verify/337: 0
corrupted double-linked list
Aborted (core dumped)
The exact keys will differ as this is a test/format run, but difference the between expected and found is consistently limited to the first character in the string.
To replicate:
mkdir build
cd build
cmake -DENABLE_LZ4=1 -DENABLE_PYTHON=1 -DENABLE_SNAPPY=1 -DENABLE_ZLIB=1 -DENABLE_ZSTD=1 -DHAVE_DIAGNOSTIC=1 -DENABLE_STRICT=1 -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -G "Ninja" ../
cd test/format
./t
We're not seeing this in other tests so it may be an issue in test/format, but marking as P2 due to reported corruptions.