Exlude perf command column in FlameGraph

[loicmathieu]

I want to create a FlameGraph based on a perf record of a GraalVM native executable. With GraalVM native executable, perf add the name of the thread to the perf report command column, this leads to un-usable FlameGraph as the frames are not merged.

I generate the perf record with perf record -F 99 -p PID sleep 10. Using DWARF symbols leads to the same issue.

Here are the first line of such a perf record:

# Overhead  Command          Shared Object                         Symbol                                                                                                                                 >
# ........  ...............  ....................................  .......................................................................................................................................>
#
     1.44%  ntloop-thread-2  [kernel.kallsyms]                     [k] acpi_pm_read
     0.91%  ecutor-thread-3  [kernel.kallsyms]                     [k] acpi_pm_read
     0.89%  ecutor-thread-5  [kernel.kallsyms]                     [k] available_idle_cpu
     0.89%  ecutor-thread-4  bookmark-service-1.0-SNAPSHOT-runner  [.] ConcurrentHashMap_get_11876bdf828aa0caf5754f7b9d3d5ee38ba37e5e
     0.80%  ecutor-thread-4  libpthread-2.31.so                    [.] __pthread_mutex_unlock
     0.80%  ntloop-thread-7  [kernel.kallsyms]                     [k] acpi_pm_read
     0.76%  ecutor-thread-4  [kernel.kallsyms]                     [k] acpi_pm_read
     0.75%  ecutor-thread-1  [kernel.kallsyms]                     [k] acpi_pm_read
     0.73%  ntloop-thread-4  [kernel.kallsyms]                     [k] do_syscall_64
     0.72%  ntloop-thread-1  [kernel.kallsyms]                     [k] acpi_pm_read
     0.71%  ecutor-thread-5  [kernel.kallsyms]                     [k] acpi_pm_read
     0.70%  ntloop-thread-3  [kernel.kallsyms]                     [k] acpi_pm_read
     0.67%  ecutor-thread-2  [kernel.kallsyms]                     [k] acpi_pm_read
     0.66%  ntloop-thread-5  [kernel.kallsyms]                     [k] acpi_pm_read
     0.64%  ecutor-thread-3  bookmark-service-1.0-SNAPSHOT-runner  [.] ArraycopySnippets_doArraycopy_043961c4b8a78c521bb6ac0bcc8243e2da5fd903
     0.61%  ecutor-thread-1  bookmark-service-1.0-SNAPSHOT-runner  [.] EnhancedQueueExecutor$PoolThreadNode_park_2242a925ee072f4927601112a1d252fa4b9f4e5f
     0.57%  ecutor-thread-6  bookmark-service-1.0-SNAPSHOT-runner  [.] ContextImpl_getEventLoop_a48ceb79cef1cc171a564acaeacaba0bcb647a3d
     0.56%  ecutor-thread-3  bookmark-service-1.0-SNAPSHOT-runner  [.] MultiThreadedMonitorSupport_getOrCreateMonitor_2ecf5995a7a109dacede518d33424ec5ebddfde6
     0.54%  ecutor-thread-1  libpthread-2.31.so                    [.] __libc_write

I don't know if it's a bug on perf, but as the command is different for each thread, the stack are not merged by the FlameGraph tools.

Is there a way to not take the command into accound when creating a FlameGraph ? Should I pre-process the perf record to update the column to the same value ?