|
NeoGraph 0.10.0
A C++17 Graph Agent Engine Library — LangGraph for C++
|
Successful node writes recorded within an in-progress super-step. More...
#include <checkpoint.h>
Public Attributes | |
| json | command |
| Serialized optional Command, or null if the node didn't emit one. | |
| std::string | node_name |
| Node that produced these writes. | |
| json | sends |
| Serialized Send vector (json array of {target_node, input}); empty if none. | |
| int64_t | step |
| Super-step number this write belongs to. | |
| std::string | task_id |
| Deterministic per-execution ID (survives replay). | |
| std::string | task_path |
| Human-readable path, e.g. "s3:executor_2" or "s3:send[0]:searcher". | |
| int64_t | timestamp |
| Unix epoch milliseconds at record time. | |
| json | writes |
| Serialized ChannelWrite vector (json array of {channel, value}). | |
Successful node writes recorded within an in-progress super-step.
PendingWrite is the fine-grained progress log that lets NeoGraph resume a partially completed super-step after a crash without re-executing nodes that already succeeded. One PendingWrite corresponds to one successful node execution; the engine records it immediately after the node returns, before applying the writes to the shared GraphState.
On resume, the engine loads all pending writes attached to the parent checkpoint, replays them into GraphState, and skips any task whose deterministic task_id is already present — so partial fan-out failures only cost the failed node's re-execution, not its successful siblings.
Definition at line 140 of file checkpoint.h.