You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When the option delete_outside_domain is enabled, Fluidity attempts to write particles after deleting their detectors, leading to write errors. In the (contrived) case of a particle subgroup with zero remaining particles after deletion, this can also crash the simulation, causing runaway memory usage. The relevant sections of the code are:
It seems like there are two parts to a potential solution: flag when detectors have been deleted and propagate this information to particles to stop the output, and add a final write before deleting the detectors. Alternatively, a simpler way would be to add a catch to prevent writing anything in the zero particles case, I'm not quite sure if that would be enough.
Hopefully the following options tree and mesh will reproduce the issue: zero_particles.tar.gz
The example options tree contains two particle subgroups, one with initialise_during_simulation set, which acts as a workaround for the issue. I get a fluidity.err-0 that contains:
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#49/x' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#49/y' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#49/z' failed.
[proc 0] E: h5pt_writedata_i4_: Write to dataset '/Step#49/id' failed.
[proc 0] E: h5pt_writedata_i4_: Write to dataset '/Step#49/proc_id' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#49/test1' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#50/x' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#50/y' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#50/z' failed.
[proc 0] E: h5pt_writedata_i4_: Write to dataset '/Step#50/id' failed.
[proc 0] E: h5pt_writedata_i4_: Write to dataset '/Step#50/proc_id' failed.
[proc 0] E: h5pt_writedata_r8_: Write to dataset '/Step#50/test1' failed.
The log shows that for the last two timesteps, the first subgroup has 0 local and 0 global detectors after moving.
The text was updated successfully, but these errors were encountered:
When the option
delete_outside_domain
is enabled, Fluidity attempts to write particles after deleting their detectors, leading to write errors. In the (contrived) case of a particle subgroup with zero remaining particles after deletion, this can also crash the simulation, causing runaway memory usage. The relevant sections of the code are:It seems like there are two parts to a potential solution: flag when detectors have been deleted and propagate this information to particles to stop the output, and add a final write before deleting the detectors. Alternatively, a simpler way would be to add a catch to prevent writing anything in the zero particles case, I'm not quite sure if that would be enough.
Hopefully the following options tree and mesh will reproduce the issue:
zero_particles.tar.gz
The example options tree contains two particle subgroups, one with
initialise_during_simulation
set, which acts as a workaround for the issue. I get afluidity.err-0
that contains:The log shows that for the last two timesteps, the first subgroup has
0 local and 0 global detectors
after moving.The text was updated successfully, but these errors were encountered: