Memory leak when using Gadget2?

[ReinosoB]

I am experiencing a memory leak when I change the internal energy of sph particles in Gadget2.
This does not occur when I do the same but using Fi. Any idea of why does this happen?
The next lines reproduces the issue:

import numpy
from amuse.units import units, constants, nbody_system
from amuse.community.gadget2.interface import Gadget2
from amuse.ext.molecular_cloud import molecular_cloud

numpy.random.seed(12122019)

Nsph =1048576
Mgas =1.e4    |units.MSun                  
Rgas =1.5     |units.parsec               

dt   =   5.   |units.yr
t_end=1000.   |units.yr  

gamma=5./3.
meanmwt=1.82  |units.amu

converter = nbody_system.nbody_to_si(Mgas, Rgas)

T_ini =200.|units.K
ratio1=5. * constants.kB * T_ini * Rgas/(3. * Mgas * (gamma-1.) * constants.G * meanmwt)
ratio2=0.0
gas = molecular_cloud(nf=64,power=1.8,targetN=Nsph, convert_nbody=converter,
                          ethep_ratio=ratio1, ekep_ratio=ratio2).result

##----sph code-------------------------------------
sph = Gadget2(converter,number_of_workers=30)
sph.initialize_code()
sph.gas_particles.add_particles(gas) 

while sph.model_time<t_end:
  sph.gas_particles.u*=0.9
  sph.evolve_model(sph.model_time+dt)

sph.stop()

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed in 28 days if no further activity occurs. Thank you for your contributions.

[rieder]

This was not picked up on at the time - but I can at least now confirm it is (still) an issue with gadget2.
We need to look into this.

[rieder]

It seems the evolve step isn't needed to experience this memory leak - even changing a property like u already does this.
I suspect this is a bug in the interface.

[rieder]

There are a lot of memory allocating calls in the gadget2 interface - I suspect one or more of these is responsible.
First thing to check would be the commit_particles function, which is also called from recommit_particles - which may not be the best idea.

[rieder]

I had a feeling of deja vu with this issue... https://groups.google.com/g/amusecode/c/Vs2lnLXkRB8
It's and issue that has been around for a while.

[rieder]

I've checked that, as in the thread above, the issue only appears when using more than one worker, with the first worker not experiencing any memory leak.