From b4746f6d5b8ebbe1267e526678a30d0396238021 Mon Sep 17 00:00:00 2001
From: Stephane Del Pino <stephane.delpino44@gmail.com>
Date: Wed, 5 Jun 2024 08:24:21 +0200
Subject: [PATCH] Remove deprecated checkpoints when not resuming from last
checkpoint
This happens when manually changing the resuming checkpoint.
Let n be the number of the last checkpoint and i be the number of the
resuming checkpoint.
If i!=n, then the checkpoints i+1, i+2, ... and n are removed when one
writes the checkpoint i+1.
---
src/utils/checkpointing/Checkpoint.cpp | 24 +++++++++++++++++-------
src/utils/checkpointing/Resume.cpp | 2 +-
2 files changed, 18 insertions(+), 8 deletions(-)
diff --git a/src/utils/checkpointing/Checkpoint.cpp b/src/utils/checkpointing/Checkpoint.cpp
index e9fe8ccd2..ffd8c1c46 100644
--- a/src/utils/checkpointing/Checkpoint.cpp
+++ b/src/utils/checkpointing/Checkpoint.cpp
@@ -51,6 +51,20 @@ checkpoint()
<< rang::fgB::yellow << ASTExecutionStack::getInstance().currentNode().begin().line << rang::fg::reset
<< " [using " << rang::fgB::cyan << checkpoint_name << rang::fg::reset << "]\n";
+ if (checkpoint_number > 0) {
+ uint64_t resuming_checkpoint_number =
+ file.getGroup("resuming_checkpoint").getAttribute("checkpoint_number").read<uint64_t>();
+ uint64_t last_checkpoint_number =
+ file.getGroup("last_checkpoint").getAttribute("checkpoint_number").read<uint64_t>();
+ for (uint64_t i = resuming_checkpoint_number + 1; i <= last_checkpoint_number; ++i) {
+ std::string old_checkpoint_name = "checkpoint_" + std::to_string(i);
+ std::cout << " - removing deprecated " << rang::fgB::yellow << old_checkpoint_name << rang::fg::reset << '\n';
+ if (file.exist(old_checkpoint_name)) {
+ file.unlink(old_checkpoint_name);
+ }
+ }
+ }
+
HighFive::Group checkpoint = file.createGroup(checkpoint_name);
uint64_t checkpoint_id =
@@ -68,6 +82,9 @@ checkpoint()
checkpoint.createAttribute("id", checkpoint_id);
checkpoint.createAttribute("data.pgs", ASTExecutionStack::getInstance().fileContent());
+ checkpoint.createAttribute("checkpoint_number", checkpoint_number);
+ ++checkpoint_number;
+
{
HighFive::Group global_variables_group = checkpoint.createGroup("singleton/global_variables");
global_variables_group.createAttribute("connectivity_id", GlobalVariableManager::instance().getConnectivityId());
@@ -209,13 +226,6 @@ checkpoint()
file.unlink("resuming_checkpoint");
}
file.createHardLink("resuming_checkpoint", checkpoint);
-
- if (file.hasAttribute("checkpoint_number")) {
- file.deleteAttribute("checkpoint_number");
- }
- file.createAttribute("checkpoint_number", checkpoint_number);
-
- ++checkpoint_number;
}
catch (HighFive::Exception& e) {
throw NormalError(e.what());
diff --git a/src/utils/checkpointing/Resume.cpp b/src/utils/checkpointing/Resume.cpp
index a4991f10e..dbb539771 100644
--- a/src/utils/checkpointing/Resume.cpp
+++ b/src/utils/checkpointing/Resume.cpp
@@ -37,7 +37,7 @@ resume()
ResumingManager& resuming_manager = ResumingManager::getInstance();
- resuming_manager.checkpointNumber() = file.getAttribute("checkpoint_number").read<uint64_t>() + 1;
+ resuming_manager.checkpointNumber() = checkpoint.getAttribute("checkpoint_number").read<uint64_t>() + 1;
std::cout << " * " << rang::fgB::green << "Resuming " << rang::fg::reset << "execution at line "
<< rang::fgB::yellow << p_node->begin().line << rang::fg::reset << " [using " << rang::fgB::cyan
--
GitLab