I tried adding
<DBRoot1>
<PreallocSpace>OFF</PreallocSpace>
</DBRoot1>

But it still breaks.

Here's some debug logs
2020-02-13 17:42:59 (7800) INFO2 : Initializing import: Table-tpch1.orders2...
2020-02-13 17:42:59 (7800) INFO2 : Initializing import: Table-tpch1.orders2; Col-o_orderkey; OID-45757; hwm-740; file-/var/lib/columnstore/data1/000.dir/000.dir/178.dir/189.dir/000.dir/FILE000.cdf
2020-02-13 17:42:59 (7800) INFO2 : Init raw data offsets in column file OID-45757; DBRoot-1; part-0; seg-0; begByte-6062080; endByte-6037504; availFileSize--24576
2020-02-13 17:42:59 (7800) INFO2 : Initializing import: Table-tpch1.orders2; Col-o_orderstatus; OID-45758; hwm-185; file-/var/lib/columnstore/data1/000.dir/000.dir/178.dir/190.dir/000.dir/FILE000.cdf
2020-02-13 17:42:59 (7800) INFO2 : Initializing import: Table-tpch1.orders2; Col-o_totalprice; OID-45759; hwm-1480; file-/var/lib/columnstore/data1/000.dir/000.dir/178.dir/191.dir/000.dir/FILE000.cdf
2020-02-13 17:42:59 (7800) INFO2 : Init raw data offsets in column file OID-45759; DBRoot-1; part-0; seg-0; begByte-12124160; endByte-12066816; availFileSize--57344
2020-02-13 17:42:59 (7800) INFO2 : Initializing import: Table-tpch1.orders2; Col-o_orderpriority; OID-45760; hwm-1480; file-/var/lib/columnstore/data1/000.dir/000.dir/178.dir/192.dir/000.dir/FILE000.cdf
2020-02-13 17:42:59 (7800) INFO2 : Opening existing store file for o_orderpriority; OID-45761; DBRoot-1; part-0; seg-0; hwm-0; file-/var/lib/columnstore/data1/000.dir/000.dir/178.dir/193.dir/000.dir/FILE000.cdf
2020-02-13 17:42:59 (7800) INFO2 : Init raw data offsets in column file OID-45760; DBRoot-1; part-0; seg-0; begByte-12124160; endByte-12066816; availFileSize--57344

2020-02-13 17:42:59 (7800) INFO : PreProcessing check completed
2020-02-13 17:42:59 (7800) INFO : preProcess completed, run time for this step : 0.0147669 seconds
2020-02-13 17:42:59 (7800) INFO : No of Read Threads Spawned = 1
2020-02-13 17:42:59 (7800) INFO : No of Parse Threads Spawned = 3
2020-02-13 17:42:59 (7800) INFO : Reading input from STDIN to import into table tpch1.orders2...
2020-02-13 17:42:59 (7800) INFO2 : Start reading and loading table tpch1.orders2
2020-02-13 17:43:00 (7800) INFO2 : Dctnry rowsExtentCheck (filling extent): OID-45760; DBRoot-1; part-0; seg-0; spaceRequired-57344; dataInBuffer-0; availSpace-57344; fColWidth-8
2020-02-13 17:43:00 (7800) INFO2 : Dctnry rowsExtentCheck: OID-45760; DBRoot-1; part-0; seg-0; Changing nRows from 30804 to -7168

Notice that begByte (read HWM) is greater than endbyte (read filesize). This causes ColumnBufferManagerDctnry::rowsExtentCheck to calculate a negative number for rows that will fit (spaceRequired = -57344) and then everything goes downhill from there. I think for non-dictionary columns, this may not be a problem as ColumnBufferManager::rowsExtentCheck() is a no-op.

So the questions are:

1. How does hwm (times blocksize 8192) become > filesize? Seems to me that's a bit alarming?
2. Why does dictionary columns use the difference here when non-dict columns don't care?
3. How come <PreallocSpace>OFF</PreallocSpace> didn't stop the problem? Is the problem new or old?
4. Does this affect compressed files?
5. What to do?

1. Preallocation code adds extra block of EMPTY values in this particular case into non-compressed files.
2. Didn't quite understand the question.
3. It should be How come <PreallocSpace>ON</PreallocSpace> not OFF.
4. No, it doesn't.
5. It is a very good question. Nothing code-wise. We should document this behavior.

Build verified: 1.4.3-1 source

engine commit:
1a65d34

Executed regression test001 test. BUG4947 passed.