myxce/src/controller/testsuite/test_waldump.c

#define _GNU_SOURCE
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>

#include <zlib.h>

#include <unity/unity.h>

#include "config.h"        /* WAL_HDR_CRC_SIZE, WAL_CRC_NONE */
#include "controller.capnp.h" /* Job encode for the JOB_SUBMIT decode test */
#include "wal_types.h"
#include "waldump.h"

/* ------------------------------------------------------------------ *
 * Test harness: hand-write raw WAL segments so the test fully controls
 * LSNs (to inject gaps), the CRC chain (to inject a break), and a real
 * Cap'n Proto Job payload (to exercise the submit jobid decode). This
 * also pins the on-disk WAL format independently of persist.c.
 * ------------------------------------------------------------------ */

static char g_dir[256];

/* Append one entry to `fp`, chaining prev_crc from *last_crc and updating it.
 * Mirrors persist.c's append: CRC32 over WAL_HDR_CRC_SIZE header bytes + payload. */
static void put_entry(FILE* fp, uint64_t term, uint64_t lsn, uint32_t type,
                      const void* payload, uint32_t plen, uint32_t* last_crc,
                      int first_in_seg)
{
    wal_entry_hdr_t hdr;
    memset(&hdr, 0, sizeof(hdr));
    hdr.term        = term;
    hdr.lsn         = lsn;
    hdr.timestamp   = 0;
    hdr.type        = type;
    hdr.payload_len = plen;
    hdr.prev_crc    = first_in_seg ? WAL_CRC_NONE : *last_crc;

    uint32_t c = crc32(0L, (const Bytef*)&hdr, WAL_HDR_CRC_SIZE);
    if (plen > 0)
        c = crc32(c, (const Bytef*)payload, plen);
    hdr.crc = c;
    *last_crc = c;

    fwrite(&hdr, sizeof(hdr), 1, fp);
    if (plen > 0)
        fwrite(payload, plen, 1, fp);
}

/* Encode a minimal Cap'n Proto Job carrying `jobid`+`subtime` (what a real
 * WAL_JOB_SUBMIT payload looks like). Returns malloc'd bytes; caller frees.
 * subtime lets a test make two payloads byte-identical (same subtime) or
 * distinct (different subtime) to exercise bug-845 dup detection. */
static uint8_t* encode_job(uint64_t jobid, uint64_t subtime, size_t* out_len)
{
    struct capn c;
    capn_init_malloc(&c);
    capn_ptr  root = capn_root(&c);
    Job_ptr   jp   = new_Job(root.seg);
    struct Job cj;
    memset(&cj, 0, sizeof(cj));
    cj.jobid   = jobid;
    cj.subtime = subtime;
    cj.state   = 1;
    write_Job(&cj, jp);
    capn_setp(root, 0, jp.p);

    size_t   sz  = capn_size(&c) + 64;
    uint8_t* buf = malloc(sz);
    int64_t  n   = capn_write_mem(&c, buf, sz, 0);
    capn_free(&c);
    *out_len = (size_t)n;
    return buf;
}

static const char* seg_path(uint64_t start_lsn, char* out, size_t n)
{
    snprintf(out, n, "%s/seg-%016lx.wal", g_dir, (unsigned long)start_lsn);
    return out;
}

/* setUp/tearDown are owned by the shared controller_test runner, so each
 * test mints its own fresh temp WAL dir here instead. */
static void fresh_dir(void)
{
    strcpy(g_dir, "/tmp/xce_waldump_XXXXXX");
    TEST_ASSERT_NOT_NULL(mkdtemp(g_dir));
}

/* ------------------------------------------------------------------ */
/* A healthy, contiguous log from LSN 1: no gaps, tip == last entry.    */
/* ------------------------------------------------------------------ */
static void test_contiguous_log_has_no_gaps(void)
{
    fresh_dir();
    char path[512];
    FILE* fp = fopen(seg_path(1, path, sizeof(path)), "wb");
    TEST_ASSERT_NOT_NULL(fp);

    uint32_t last = WAL_CRC_NONE;
    put_entry(fp, 0, 1, WAL_NOOP, NULL, 0, &last, 1);
    size_t   p2, p3;
    uint8_t* j2 = encode_job(1001, 100, &p2); /* distinct jobs → no dup */
    uint8_t* j3 = encode_job(1002, 100, &p3);
    put_entry(fp, 0, 2, WAL_JOB_SUBMIT, j2, (uint32_t)p2, &last, 0);
    put_entry(fp, 0, 3, WAL_JOB_SUBMIT, j3, (uint32_t)p3, &last, 0);
    free(j2);
    free(j3);
    fclose(fp);

    waldump_stats_t st;
    int rc = waldump_path(g_dir, stdout, &st);
    TEST_ASSERT_EQUAL_INT(0, rc);
    TEST_ASSERT_EQUAL_UINT(3, st.n_entries);
    TEST_ASSERT_EQUAL_UINT(2, st.n_submits);
    TEST_ASSERT_EQUAL_UINT(0, st.n_gaps);
    TEST_ASSERT_EQUAL_UINT(0, st.n_dup_submits); /* distinct jobids */
    TEST_ASSERT_EQUAL_UINT64(1, st.first_lsn);
    TEST_ASSERT_EQUAL_UINT64(3, st.last_lsn);
    TEST_ASSERT_EQUAL_UINT64(3, st.recovered_tip);
    TEST_ASSERT_EQUAL_INT(0, st.crc_broken);
}

/* ------------------------------------------------------------------ */
/* The failover bug signature: log starts at a high LSN (empty-rejoin   */
/* caught up from the watermark) so the committed prefix is missing.    */
/* ------------------------------------------------------------------ */
static void test_missing_prefix_flagged_as_gap(void)
{
    fresh_dir();
    char path[512];
    /* seg file named from its first LSN, 13 (like ctrl-3 in the live repro) */
    FILE* fp = fopen(seg_path(13, path, sizeof(path)), "wb");
    TEST_ASSERT_NOT_NULL(fp);

    uint32_t last = WAL_CRC_NONE;
    put_entry(fp, 0, 13, WAL_NOOP, NULL, 0, &last, 1);
    size_t   p14, p15;
    uint8_t* j14 = encode_job(2001, 100, &p14);
    uint8_t* j15 = encode_job(2002, 100, &p15);
    put_entry(fp, 0, 14, WAL_JOB_SUBMIT, j14, (uint32_t)p14, &last, 0);
    put_entry(fp, 0, 15, WAL_JOB_SUBMIT, j15, (uint32_t)p15, &last, 0);
    free(j14);
    free(j15);
    fclose(fp);

    waldump_stats_t st;
    int rc = waldump_path(g_dir, stdout, &st);
    TEST_ASSERT_EQUAL_INT(0, rc);
    TEST_ASSERT_EQUAL_UINT(3, st.n_entries);
    TEST_ASSERT_EQUAL_UINT64(13, st.first_lsn);
    /* prefix [1..12] absent → exactly one gap */
    TEST_ASSERT_EQUAL_UINT(1, st.n_gaps);
}

/* ------------------------------------------------------------------ */
/* bug-845: each job double-appended (same jobid at adjacent LSN). The  */
/* identical-payload case = a literal byte-for-byte double-write.       */
/* ------------------------------------------------------------------ */
static void test_duplicate_submit_detection(void)
{
    fresh_dir();
    char path[512];
    FILE* fp = fopen(seg_path(1, path, sizeof(path)), "wb");
    TEST_ASSERT_NOT_NULL(fp);

    uint32_t last = WAL_CRC_NONE;
    put_entry(fp, 0, 1, WAL_NOOP, NULL, 0, &last, 1);
    /* Same jobid + same subtime → byte-identical payload (literal dup). */
    size_t   pa, pb;
    uint8_t* ja = encode_job(7001, 555, &pa);
    uint8_t* jb = encode_job(7001, 555, &pb);
    put_entry(fp, 0, 2, WAL_JOB_SUBMIT, ja, (uint32_t)pa, &last, 0);
    put_entry(fp, 0, 3, WAL_JOB_SUBMIT, jb, (uint32_t)pb, &last, 0);
    /* A second job, also doubled but with a DIFFERING subtime per copy. */
    size_t   pc, pd;
    uint8_t* jc = encode_job(7002, 600, &pc);
    uint8_t* jd = encode_job(7002, 601, &pd);
    put_entry(fp, 0, 4, WAL_JOB_SUBMIT, jc, (uint32_t)pc, &last, 0);
    put_entry(fp, 0, 5, WAL_JOB_SUBMIT, jd, (uint32_t)pd, &last, 0);
    free(ja); free(jb); free(jc); free(jd);
    fclose(fp);

    waldump_stats_t st;
    int rc = waldump_path(g_dir, stdout, &st);
    TEST_ASSERT_EQUAL_INT(0, rc);
    TEST_ASSERT_EQUAL_UINT(4, st.n_submits);
    TEST_ASSERT_EQUAL_UINT(2, st.n_dup_submits);   /* both pairs flagged */
    TEST_ASSERT_EQUAL_UINT(1, st.n_dup_identical); /* only the first is byte-identical */
    TEST_ASSERT_EQUAL_UINT(0, st.n_gaps);
}

/* ------------------------------------------------------------------ */
/* An interior hole (a skipped LSN between two present entries).        */
/* ------------------------------------------------------------------ */
static void test_interior_gap_flagged(void)
{
    fresh_dir();
    /* Two segments: [1..2] then [5..6]; 3,4 are missing. */
    char p1[512], p2[512];
    FILE* f1 = fopen(seg_path(1, p1, sizeof(p1)), "wb");
    uint32_t last = WAL_CRC_NONE;
    put_entry(f1, 0, 1, WAL_NOOP, NULL, 0, &last, 1);
    put_entry(f1, 0, 2, WAL_JOB_UPDATE, "x", 1, &last, 0);
    fclose(f1);

    FILE* f2 = fopen(seg_path(5, p2, sizeof(p2)), "wb");
    uint32_t last2 = WAL_CRC_NONE;
    put_entry(f2, 0, 5, WAL_JOB_UPDATE, "y", 1, &last2, 1);
    put_entry(f2, 0, 6, WAL_JOB_UPDATE, "z", 1, &last2, 0);
    fclose(f2);

    waldump_stats_t st;
    int rc = waldump_path(g_dir, stdout, &st);
    TEST_ASSERT_EQUAL_INT(0, rc);
    TEST_ASSERT_EQUAL_UINT(4, st.n_entries);
    TEST_ASSERT_EQUAL_UINT(2, st.n_segments);
    TEST_ASSERT_EQUAL_UINT64(1, st.first_lsn);
    TEST_ASSERT_EQUAL_UINT64(6, st.last_lsn);
    TEST_ASSERT_EQUAL_UINT(1, st.n_gaps); /* the 3..4 hole */
}

/* ------------------------------------------------------------------ */
/* A CRC break truncates the recovered tip below the last physical LSN. */
/* ------------------------------------------------------------------ */
static void test_crc_break_sets_recovered_tip(void)
{
    fresh_dir();
    char path[512];
    FILE* fp = fopen(seg_path(1, path, sizeof(path)), "wb");
    uint32_t last = WAL_CRC_NONE;
    put_entry(fp, 0, 1, WAL_NOOP, NULL, 0, &last, 1);
    put_entry(fp, 0, 2, WAL_JOB_UPDATE, "ok", 2, &last, 0);
    fclose(fp);

    /* Corrupt the second entry's stored CRC in place. */
    fp = fopen(path, "r+b");
    wal_entry_hdr_t h;
    long off2 = sizeof(wal_entry_hdr_t) + 0; /* entry 1 has 0 payload */
    fseek(fp, off2, SEEK_SET);
    fread(&h, sizeof(h), 1, fp);
    h.crc ^= 0xdeadbeefu; /* break it */
    fseek(fp, off2, SEEK_SET);
    fwrite(&h, sizeof(h), 1, fp);
    fclose(fp);

    waldump_stats_t st;
    int rc = waldump_path(g_dir, stdout, &st);
    TEST_ASSERT_EQUAL_INT(0, rc);
    TEST_ASSERT_EQUAL_UINT64(1, st.recovered_tip); /* stops before the bad LSN 2 */
    TEST_ASSERT_NOT_EQUAL(0, st.crc_broken);
}

/* ------------------------------------------------------------------ */
/* Missing path is a hard error.                                        */
/* ------------------------------------------------------------------ */
static void test_missing_path_errors(void)
{
    waldump_stats_t st;
    int rc = waldump_path("/tmp/xce_waldump_does_not_exist_zzz", stdout, &st);
    TEST_ASSERT_LESS_THAN_INT(0, rc);
}

/* ------------------------------------------------------------------ */
/* A non-WAL file (e.g. a config blob) must ERROR, not report 0 entries.*/
/* ------------------------------------------------------------------ */
static void test_non_wal_file_errors(void)
{
    fresh_dir();
    char path[512];
    snprintf(path, sizeof(path), "%s/cluster.bin", g_dir);
    FILE* fp = fopen(path, "wb");
    TEST_ASSERT_NOT_NULL(fp);
    const char junk[] = "not a wal segment — a config blob";
    fwrite(junk, sizeof(junk), 1, fp);
    fclose(fp);

    /* Pointing at a non-seg-*.wal file must be a hard error (-EINVAL), not a
     * silent "0 entries" scan. */
    waldump_stats_t st;
    int rc = waldump_path(path, stdout, &st);
    TEST_ASSERT_LESS_THAN_INT(0, rc);
}

void test_waldump(void)
{
    RUN_TEST(test_contiguous_log_has_no_gaps);
    RUN_TEST(test_missing_prefix_flagged_as_gap);
    RUN_TEST(test_duplicate_submit_detection);
    RUN_TEST(test_interior_gap_flagged);
    RUN_TEST(test_crc_break_sets_recovered_tip);
    RUN_TEST(test_missing_path_errors);
    RUN_TEST(test_non_wal_file_errors);
}