Example: two rte_channel_t links (one per "machine"), registered into a rte_voter_t and synchronized via rte_checkpoint (ADR-017, rewired onto rte_voter_t by ADR-025) instead of wall-clock agreement. More...

#include <setjmp.h>
#include <stdio.h>
#include <string.h>
#include "rte/checkpoint/rte_checkpoint.h"
#include "rte/checksum/rte_checksum.h"
#include "rte/clocksync/rte_clocksync.h"
#include "rte_osadapter/clocksync/rte_osadapter_clocksync.h"
#include "rte/safestate/rte_safestate.h"
#include "rte/voter/rte_voter.h"

Include dependency graph for geo_distributed_checkpoint_sync.c:

Macros
#define	SITE_WEST 0
#define	SITE_EAST 1
#define	SITE_COUNT 2U

Functions
static rte_status_t	mock_send (void channel, const void data, size_t data_size)
static rte_status_t	mock_recv (void channel, void data, size_t data_size, uint32_t timeout_ms)
static rte_status_t	clocksync_get_offset_ms (int64_t *out_offset_ms)
static rte_status_t	clocksync_get_quality (rte_clocksync_quality_t *out_quality)
static void	demo_safestate_handler (rte_safestate_level_t level, rte_safestate_reason_t reason, const char file, int32_t line, const char message)
int	main (void)

Variables
static int	g_site_index [SITE_COUNT] = { SITE_WEST, SITE_EAST }
static void *	g_channel_handles [SITE_COUNT] = { &g_site_index[SITE_WEST], &g_site_index[SITE_EAST] }
static rte_vital_message_t	g_mailbox [SITE_COUNT]
static bool	g_mail_ready [SITE_COUNT]
static bool	g_link_down [SITE_COUNT]
static jmp_buf	g_safestate_jmp

Detailed Description

Example: two rte_channel_t links (one per "machine"), registered into a rte_voter_t and synchronized via rte_checkpoint (ADR-017, rewired onto rte_voter_t by ADR-025) instead of wall-clock agreement.

Demonstrates the answer to "how do channels on different hardware, in different locations, get compared as if they were the same silicon": not by making their clocks agree (rte_clocksync is diagnostic only - see its header), but by having both sites confirm the same checkpoint_id to each other within a bounded timeout before the real vote happens. If a site doesn't confirm in time, that's a fault and the framework enters safe-state itself - it is not left to the application to notice.

This example simulates the network with in-memory mailboxes (one per site) instead of a real socket/serial link, so it runs standalone with no external dependencies - swap mock_backend_send/recv for a real transport (e.g. built on rte_ipc, TCP, or a serial link) to deploy this for real, per ADR-017 section 2.2 ("no new OSAdapter of its own - reuses whatever transport is already registered on each channel").

Scenario: Cycle 1: both sites confirm checkpoint 1 in time -> vote succeeds. Cycle 2: WEST's confirmation is deliberately dropped (simulating a lost/delayed message) -> checkpoint fails -> safe-state.

Compile: gcc -std=c99 -Wall -Wextra -o geo_checkpoint_sync \ geo_distributed_checkpoint_sync.c \ -I../include -L../build \ -lrte_channels -lrte_oal -lrte_core

Definition in file geo_distributed_checkpoint_sync.c.