Complete state of a dynamic event tree run at a sequence boundary. More...

#include <DETCheckpoint.h>

Public Member Functions
template<class Archive >
void	serialize (Archive &ar, unsigned int)
	Boost serialization of the archive layout. More...

void	capture_branch_center (const BranchRecord &br)
	Serialize `br` (Boost binary archive) into branch_center_blob, replacing its previous content. More...

void	restore_branch_center (BranchRecord &br) const
	Deserialize branch_center_blob into `br`. The blob must be uncompressed, as load() leaves it; a Boost archive exception propagates if the blob is empty or malformed. More...

bool	config_matches (uint64_t current) const
	Does this checkpoint's configuration digest match `current`, the digest of the running process (g_sim.config_fingerprint)? More...

void	save (const std::string &path)
	Write the checkpoint to `path`. More...

Static Public Member Functions
static bool	version_supported (int v)
	Whether load() accepts a file with stored version `v`. More...

static std::string	lz4_pack_string (const std::string &s)
	LZ4-compress `s` with the 4-byte size header of lz4_compress(). More...

static std::string	lz4_unpack_string (const std::string &s)
	Inverse of lz4_pack_string(). Exits the process (lz4_decompress) if `s` does not decompress to its recorded size. More...

static std::string	pack_rinfo (const RestartNodeInfo &ri)
	Serialize one snapshot (Boost binary archive) and LZ4-compress it. More...

static std::shared_ptr< RestartNodeInfo >	unpack_rinfo (const std::string &blob)
	Inverse of pack_rinfo(). Deserialization pre-seeds the snapshot from the process-local shared masters (SharedConfig.h); with no master registered, its configuration members stay default-valued. More...

static DETCheckpoint	load (const std::string &path)
	Read a checkpoint file. More...

Public Attributes
int	version = VERSION
	Format version stored in the archive. More...

int	sequences_completed = 0
	Sequences finished when the checkpoint was taken. More...

int	sequence_counter = 0

uint64_t	config_fingerprint = 0

bool	compress_queue = false

bool	blobs_compressed = false

std::map< int, std::vector< char > >	source_blobs
	source key -> restart blob More...

std::map< int, int >	cid_source

int	max_branch_id = -1

std::string	branch_center_blob

std::vector< std::string >	packed_rinfos
	lz4(serialized rinfo), one per unique snapshot More...

std::map< int, int >	restart_state_pack_ref
	cid -> packed index (-1 = null entry) More...

std::vector< int >	task_state_pack_ref
	task position -> packed index (-1 = null) More...

std::map< int, std::shared_ptr< RestartNodeInfo > >	restart_states
	cid -> snapshot; read by serial resume More...

std::map< int, std::string >	branch_record_blobs
	cid -> serialized BranchRecord of the holding client; filled by the MPI writer only and used there to build branch_center_blob More...

std::vector< PendingBranch >	pending_tasks
	Unexplored branches. More...

std::vector< std::shared_ptr< RestartNodeInfo > >	task_restart_states
	Snapshot per pending task (null = none); read by MPI resume. More...

std::vector< std::string >	task_branch_blobs
	Always empty; resume dispatches branch_center_blob. More...

std::vector< SequenceReport >	completed_reports
	Reports of finished sequences (filled only when the visualization server is on) More...

Static Public Attributes
static constexpr int	VERSION = 9

Detailed Description

Complete state of a dynamic event tree run at a sequence boundary.

One Boost binary archive per file. The same format is written by the serial loop (ADSIDAC.cpp) and by the MPI server (Server::write_checkpoint) and can be resumed in either mode and with any rank count. Crew snapshots inside it are state-only: a resume rebuilds configuration from the project's input files, which is why the configuration fingerprint is checked. Not thread-safe; build, save and load from one thread.

Definition at line 45 of file DETCheckpoint.h.

Member Function Documentation

◆ capture_branch_center()

void DETCheckpoint::capture_branch_center ( const BranchRecord & br )

inline

Serialize br (Boost binary archive) into branch_center_blob, replacing its previous content.

Definition at line 151 of file DETCheckpoint.h.

References branch_center_blob.

Referenced by build_serial_checkpoint().

Here is the caller graph for this function:

◆ config_matches()

bool DETCheckpoint::config_matches ( uint64_t current ) const

inline

Does this checkpoint's configuration digest match current, the digest of the running process (g_sim.config_fingerprint)?

Returns: true on equality, and also when either digest is 0 (not computed, as in unit tests) or the stored version is below 6. Real runs always carry a computed digest on both sides.

Definition at line 177 of file DETCheckpoint.h.

References config_fingerprint, and version.

◆ load()

static DETCheckpoint DETCheckpoint::load ( const std::string & path )

inlinestatic

Read a checkpoint file.

Exits the process with status 1 (message on stderr) if the file cannot be opened, the archive is corrupt or truncated, or its version is not supported; it never returns a partial object. In MPI only rank 0 calls it, and the exit is a plain std::exit(1), not MPI_Abort. On return the at-rest LZ4 layer is stripped (blobs_compressed == false), restart_states and task_restart_states are rebuilt with their shared_ptr aliasing, and the packed members are empty. When compress_queue is true, source_blobs still carry their queue-layer LZ4 encoding. Snapshot deserialization uses the shared masters of SharedConfig.h, so register them first. The configuration fingerprint is NOT checked here; callers use config_matches().

Definition at line 312 of file DETCheckpoint.h.

References blobs_compressed, branch_center_blob, branch_record_blobs, compress_queue, lz4_decompress(), lz4_unpack_string(), packed_rinfos, restart_state_pack_ref, restart_states, source_blobs, task_restart_states, task_state_pack_ref, unpack_rinfo(), VERSION, version, and version_supported().

Referenced by Server::dispatch_resume_tasks(), and main().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ lz4_pack_string()

static std::string DETCheckpoint::lz4_pack_string ( const std::string & s )

inlinestatic

LZ4-compress s with the 4-byte size header of lz4_compress().

Definition at line 183 of file DETCheckpoint.h.

References lz4_compress().

Referenced by pack_rinfo(), and save().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ lz4_unpack_string()

static std::string DETCheckpoint::lz4_unpack_string ( const std::string & s )

inlinestatic

Inverse of lz4_pack_string(). Exits the process (lz4_decompress) if s does not decompress to its recorded size.

Definition at line 189 of file DETCheckpoint.h.

References lz4_decompress().

Referenced by load(), and unpack_rinfo().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ pack_rinfo()

static std::string DETCheckpoint::pack_rinfo ( const RestartNodeInfo & ri )

inlinestatic

Serialize one snapshot (Boost binary archive) and LZ4-compress it.

Definition at line 195 of file DETCheckpoint.h.

References lz4_pack_string().

Referenced by save().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ restore_branch_center()

void DETCheckpoint::restore_branch_center ( BranchRecord & br ) const

inline

Deserialize branch_center_blob into br. The blob must be uncompressed, as load() leaves it; a Boost archive exception propagates if the blob is empty or malformed.

Definition at line 161 of file DETCheckpoint.h.

References branch_center_blob.

◆ save()

void DETCheckpoint::save ( const std::string & path )

inline

Write the checkpoint to path.

Writes <path>.tmp and renames it into place, so a process killed mid-save leaves any earlier file at path intact. MUTATES the object: source_blobs (unless compress_queue), branch_record_blobs and branch_center_blob are LZ4-compressed in place and blobs_compressed is set, so build a fresh object for every save. restart_states and task_restart_states stay usable. On failure (cannot open, short write, rename error) it prints "ERROR: ..." on stderr and returns without throwing or exiting (a short write also removes the temp file); success prints "[CHECKPOINT] Saved: ..." on stdout. There is no return value, so a caller cannot tell the two apart.

Definition at line 227 of file DETCheckpoint.h.

References blobs_compressed, branch_center_blob, branch_record_blobs, compress_queue, lz4_compress(), lz4_pack_string(), pack_rinfo(), packed_rinfos, pending_tasks, restart_state_pack_ref, restart_states, sequences_completed, source_blobs, task_restart_states, and task_state_pack_ref.

Referenced by Server::write_checkpoint().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ serialize()

template<class Archive >

void DETCheckpoint::serialize	(	Archive &	ar,
		unsigned int
	)

inline

Boost serialization of the archive layout.

load() accepts only files with version >= 9, and save() writes VERSION, so for every accepted archive each version >= N guard below is true and both version < 7 guards are false: the raw restart_states and task_restart_states containers are never archived, only their packed form. For a file with a lower stored version the guards select the fields that version carried; load() then rejects the file (or reports it corrupt if the archive cannot be parsed).

Definition at line 113 of file DETCheckpoint.h.

References blobs_compressed, branch_center_blob, branch_record_blobs, cid_source, completed_reports, compress_queue, config_fingerprint, max_branch_id, packed_rinfos, pending_tasks, restart_state_pack_ref, restart_states, sequence_counter, sequences_completed, source_blobs, task_branch_blobs, task_restart_states, task_state_pack_ref, and version.

◆ unpack_rinfo()

static std::shared_ptr< RestartNodeInfo > DETCheckpoint::unpack_rinfo ( const std::string & blob )

inlinestatic

Inverse of pack_rinfo(). Deserialization pre-seeds the snapshot from the process-local shared masters (SharedConfig.h); with no master registered, its configuration members stay default-valued.

Definition at line 206 of file DETCheckpoint.h.

References lz4_unpack_string().

Referenced by load().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ version_supported()

static bool DETCheckpoint::version_supported ( int v )

inlinestatic

Whether load() accepts a file with stored version v.

Returns: true for v >= 9. Files below that carry class layouts this binary does not parse, so load() exits instead of reading them.

Definition at line 170 of file DETCheckpoint.h.

Referenced by load().

Here is the caller graph for this function:

Member Data Documentation

◆ blobs_compressed

bool DETCheckpoint::blobs_compressed = false

At-rest compression flag. save() LZ4-compresses the blob payloads inside the archive (RELAP restart sources dominate checkpoint size) and load() strips the layer again, so consumers always see uncompressed bytes. Sources already compressed by –compress-queue keep their queue-layer bytes (owned by that flag) and are not compressed a second time.

Definition at line 67 of file DETCheckpoint.h.

Referenced by load(), save(), and serialize().

◆ branch_center_blob

std::string DETCheckpoint::branch_center_blob

Serialized merged BranchRecord. Serial resume restores it as the global tree; MPI resume sends it with every dispatched task.

Definition at line 80 of file DETCheckpoint.h.

Referenced by capture_branch_center(), load(), restore_branch_center(), save(), serialize(), and Server::write_checkpoint().

◆ branch_record_blobs

std::map<int, std::string> DETCheckpoint::branch_record_blobs

cid -> serialized BranchRecord of the holding client; filled by the MPI writer only and used there to build branch_center_blob

Definition at line 94 of file DETCheckpoint.h.

Referenced by load(), save(), serialize(), and Server::write_checkpoint().

◆ cid_source

std::map<int, int> DETCheckpoint::cid_source

pending CID -> source key

Definition at line 72 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), serialize(), and Server::write_checkpoint().

◆ completed_reports

std::vector<SequenceReport> DETCheckpoint::completed_reports

Reports of finished sequences (filled only when the visualization server is on)

Definition at line 101 of file DETCheckpoint.h.

Referenced by serialize(), and Server::write_checkpoint().

◆ compress_queue

bool DETCheckpoint::compress_queue = false

Whether the restart blobs carry the queue-layer LZ4 encoding of –compress-queue. A resume adopts this value, so the blobs are always decoded the way they were encoded. Serial checkpoints always store false.

Definition at line 61 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), load(), save(), serialize(), and Server::write_checkpoint().

◆ config_fingerprint

uint64_t DETCheckpoint::config_fingerprint = 0

FNV-1a digest of the immutable configuration inputs (deck, Input/‍**, ZCtrlPar.txt) at write time. A resume compares it with its own digest and refuses on mismatch: the snapshots are state-only, so a resume rebuilds configuration from the project files, and different files would mean a different tree. 0 = not computed (see config_matches()).

Definition at line 57 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), config_matches(), serialize(), and Server::write_checkpoint().

◆ max_branch_id

int DETCheckpoint::max_branch_id = -1

Highest branch ID issued by the checkpointed run, exactly as issued. A resumed run derives a fresh ID epoch above it so its new IDs can never collide with identities that live on inside merged PE outputs. -1 = none.

Definition at line 76 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), serialize(), and Server::write_checkpoint().

◆ packed_rinfos

std::vector<std::string> DETCheckpoint::packed_rinfos

lz4(serialized rinfo), one per unique snapshot

Definition at line 88 of file DETCheckpoint.h.

Referenced by load(), save(), and serialize().

◆ pending_tasks

std::vector<PendingBranch> DETCheckpoint::pending_tasks

Unexplored branches.

Definition at line 97 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), save(), serialize(), and Server::write_checkpoint().

◆ restart_state_pack_ref

std::map<int, int> DETCheckpoint::restart_state_pack_ref

cid -> packed index (-1 = null entry)

Definition at line 89 of file DETCheckpoint.h.

Referenced by load(), save(), and serialize().

◆ restart_states

std::map<int, std::shared_ptr<RestartNodeInfo> > DETCheckpoint::restart_states

cid -> snapshot; read by serial resume

Definition at line 93 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), load(), save(), serialize(), and Server::write_checkpoint().

◆ sequence_counter

int DETCheckpoint::sequence_counter = 0

Next global sequence ID to assign

Definition at line 51 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), serialize(), and Server::write_checkpoint().

◆ sequences_completed

int DETCheckpoint::sequences_completed = 0

Sequences finished when the checkpoint was taken.

Definition at line 50 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), save(), serialize(), and Server::write_checkpoint().

◆ source_blobs

std::map<int, std::vector<char> > DETCheckpoint::source_blobs

source key -> restart blob

Definition at line 71 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), load(), save(), serialize(), and Server::write_checkpoint().

◆ task_branch_blobs

std::vector<std::string> DETCheckpoint::task_branch_blobs

Always empty; resume dispatches branch_center_blob.

Definition at line 99 of file DETCheckpoint.h.

Referenced by serialize().

◆ task_restart_states

std::vector<std::shared_ptr<RestartNodeInfo> > DETCheckpoint::task_restart_states

Snapshot per pending task (null = none); read by MPI resume.

Definition at line 98 of file DETCheckpoint.h.

Referenced by build_serial_checkpoint(), load(), save(), serialize(), and Server::write_checkpoint().

◆ task_state_pack_ref

std::vector<int> DETCheckpoint::task_state_pack_ref

task position -> packed index (-1 = null)

Definition at line 90 of file DETCheckpoint.h.

Referenced by load(), save(), and serialize().

◆ VERSION

constexpr int DETCheckpoint::VERSION = 9

staticconstexpr

Current file format version. save() always writes it; load() refuses any file whose stored version is lower (see version_supported()).

Definition at line 48 of file DETCheckpoint.h.

Referenced by load().

◆ version

int DETCheckpoint::version = VERSION

Format version stored in the archive.

Definition at line 49 of file DETCheckpoint.h.

Referenced by config_matches(), load(), and serialize().

The documentation for this struct was generated from the following file:

src/ADSIDAC/Global/DETCheckpoint.h

Public Member Functions

Static Public Member Functions

Public Attributes

Static Public Attributes

Detailed Description

Member Function Documentation

◆ capture_branch_center()

◆ config_matches()

◆ load()

◆ lz4_pack_string()

◆ lz4_unpack_string()

◆ pack_rinfo()

◆ restore_branch_center()

◆ save()

◆ serialize()

◆ unpack_rinfo()

◆ version_supported()

Member Data Documentation

◆ blobs_compressed

◆ branch_center_blob

◆ branch_record_blobs

◆ cid_source

◆ completed_reports

◆ compress_queue

◆ config_fingerprint

◆ max_branch_id

◆ packed_rinfos

◆ pending_tasks

◆ restart_state_pack_ref

◆ restart_states

◆ sequence_counter

◆ sequences_completed

◆ source_blobs

◆ task_branch_blobs

◆ task_restart_states

◆ task_state_pack_ref

◆ VERSION

◆ version