Kernel Restarting - src/tcmalloc.cc:283] Attempt to free invalid pointer

[wang3t]

Q1: Any suggestion to resolve this ?

Using Jupyter notebook with two tutorial on macOS Sierra(10.12.5)

1. Basics.ipynb
2. Tutorial_3_MNIST.ipynb
   Having same/similar issues Kernel Restarting with similar MSG at different address
   src/tcmalloc.cc:283] Attempt to free invalid pointer

For Tutorial_3_MNIST, it contains more info
src/tcmalloc.cc:283] Attempt to free invalid pointer 0x1004b4270
*** Aborted at 1496414515 (unix time) try "date -d @1496414515" if you are using GNU date ***
PC: @ 0x7fff8d79ad42 __pthread_kill
*** SIGABRT (@0x7fff8d79ad42) received by PID 82928 (TID 0x700007183000) stack trace: ***
@ 0x7fff8d87bb3a _sigtramp
@ 0x7000071805c0 (unknown)
@ 0x7fff8d700420 abort
@ 0x10f7e4369 tcmalloc::Log()
@ 0x10f7f41f3 tc_free
@ 0x104685bc2 sqlite3MemFree
@ 0x10468700f sqlite3_free
@ 0x10468715a sqlite3DbFree
@ 0x1046c1b64 vdbeMemClear
@ 0x1046c1bc9 sqlite3VdbeMemRelease
@ 0x1046cd945 vdbeUnbind
@ 0x1046cda19 bindText
@ 0x1046cdea7 sqlite3_bind_text
@ 0x10055d1d3 pysqlite_statement_bind_parameter
@ 0x10055d4e6 pysqlite_statement_bind_parameters
@ 0x10055ba7b _pysqlite_query_execute
@ 0x10000c761 PyObject_Call
@ 0x1000a1874 PyEval_CallObjectWithKeywords
@ 0x1005590a5 pysqlite_connection_execute
@ 0x1000a3302 PyEval_EvalFrameEx
@ 0x1000ac7dc PyEval_EvalCodeEx
@ 0x1000ac988 fast_function
@ 0x1000a338a PyEval_EvalFrameEx
@ 0x1000ac7dc PyEval_EvalCodeEx
@ 0x100035c1b function_call
@ 0x10000c761 PyObject_Call
@ 0x1000a3cea PyEval_EvalFrameEx
@ 0x1000ac7dc PyEval_EvalCodeEx
@ 0x1000ac988 fast_function
@ 0x1000a338a PyEval_EvalFrameEx
@ 0x1000ac7dc PyEval_EvalCodeEx
@ 0x1000ac988 fast_function
[I 07:41:57.840 NotebookApp] KernelRestarter: restarting kernel (1/5)

[wang3t]

// 1. re-build without LEVELDB,
$ cmake -DUSE_CUDA=OFF -DUSE_LEVELDB=OFF ..

// 2. Following tutorials work !

• Basics.ipynb
• Tutorial_3_MNIST.ipynb

[aaronmarkham]

Hi @wang3t, this seems to work to get past the leveldb/tcmalloc problem on MacOS, but then you can't run the main MNIST tutorial which uses LevelDB. It could be possible though to switch the database type to minidb and run the tutorial without LevelDB until this tcmalloc problem is sorted out.
/usr/local/binaries/make_mnist_db --channel_first --db leveldb... that would be one change there on the --db flag plus cleaning up the db naming to reflect minidb instead.

[moderato]

@wang3t I got the same bug when I was using matcaffe. Using this method and rebuilding caffe works for me. Thanks!

[aywi]

@wang3t This works for matcaffe too. Thanks!