From 3d5c3d41a4dc9bec24cbbf564f380eaca9c82ca3 Mon Sep 17 00:00:00 2001 From: Andrew Coughtrie <24609575+andrewcoughtrie@users.noreply.github.com> Date: Thu, 5 Feb 2026 10:44:45 +0000 Subject: [PATCH] Update checkpointing to include new functionality --- .../checkpointing/lfric_atm_checkpoint.rst | 18 +++++++++++++----- 1 file changed, 13 insertions(+), 5 deletions(-) diff --git a/documentation/source/user_guide/checkpointing/lfric_atm_checkpoint.rst b/documentation/source/user_guide/checkpointing/lfric_atm_checkpoint.rst index 17be30420..2fee14cbd 100644 --- a/documentation/source/user_guide/checkpointing/lfric_atm_checkpoint.rst +++ b/documentation/source/user_guide/checkpointing/lfric_atm_checkpoint.rst @@ -9,8 +9,8 @@ LFRic atmosphere checkpoint/restart system ========================================== -The LFRic atmosphere ``lfric_atm`` application can be configured to generate a -checkpoint dump at the end of each model run. The checkpoint dump can be read in +The LFRic atmosphere ``lfric_atm`` application can be configured to generate +checkpoint dumps at multiple points within a model run. The checkpoint dump can be read in by a new integration of the model allowing further timesteps to be run. The dump is written using XIOS. @@ -18,9 +18,17 @@ Requesting checkpoint restart ----------------------------- Set ``checkpoint_write=.true.`` in the ``io`` namelist of the model -configuration to generate a checkpoint dump at the end of a model run. The -checkpoint dump will be named after the ``checkpoint_stem_name`` string in the -``files`` namelist appended with the number of the last timestep of the run. +configuration to turn on checkpoint writing. The times when checkpoint files +will be written are defined by the ``checkpoint_times`` list. The +``checkpoint_times`` must be real values in seconds corresponding to an integer +number of timesteps in the model run (e.g. if ``dt=0.5`` the +``checkpoint_times`` can include ``0.5`` and ``1.0`` etc. but **not** ``0.3``). +Setting ``end_of_run_checkpoint=.true.`` will write a checkpoint file at the end +of the run. Only one checkpoint file will be written at the end of a run even if +both the final timestep is included in the ``checkpoint_times`` list and the +``end_of_run_checkpoint`` flag is set to ``.true.``. The checkpoint dumps will +be named after the ``checkpoint_stem_name`` string in the ``files`` namelist +appended with timestep number of the checkpoint time. Set ``checkpoint_read=.true.`` in the ``io`` namelist to restart a run from an existing checkpoint dump. The expected start timestep will be defined by