mdbx-tools: rework mdbx_reader_list() & mdbx_stat.

Change-Id: I0524cad93ca439e74eba9486cbcbeacf4253dd84

mdbx.h

@@ -2716,31 +2716,49 @@ LIBMDBX_API int mdbx_cmp(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *a,
 LIBMDBX_API int mdbx_dcmp(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *a,
                           const MDBX_val *b);
 
-/* A callback function used to print a message from the library.
+/* A callback function used to enumerate the reader lock table.
  *
- * [in] msg   The string to be printed.
+ * [in] ctx           An arbitrary context pointer for the callback.
- * [in] ctx   An arbitrary context pointer for the callback.
+ * [in] num           The serial number during enumeration, starting from 1.
+ * [in] slot          The reader lock table slot number.
+ * [in] txnid         The ID of the transaction being read,
+ *                    i.e. the MVCC-snaphot number.
+ * [in] lag           The lag from a recent MVCC-snapshot, i.e. the number of
+ *                    committed transaction since read transaction started.
+ * [in] pid           The reader process ID.
+ * [in] thread        The reader thread ID.
+ * [in] bytes_used    The number of last used page in the MVCC-snapshot which
+ *                    being read, i.e. database file can't shrinked beyond this.
+ * [in] bytes_retired The total size of the database pages that were retired by
+ *                    committed write transactions after the reader's
+ *                    MVCC-snapshot, i.e. the space which would be freed after
+ *                    the Reader releases the MVCC-snapshot for reuse by
+ *                    completion read transaction.
  *
  * Returns < 0 on failure, >= 0 on success. */
-typedef int(MDBX_msg_func)(const char *msg, void *ctx);
+typedef int(MDBX_reader_list_func)(void *ctx, int num, int slot, mdbx_pid_t pid,
+                                   mdbx_tid_t thread, uint64_t txnid,
+                                   uint64_t lag, size_t bytes_used,
+                                   size_t bytes_retired);
 
-/* FIXME: Rework this function.
+/* Enumarete the entries in the reader lock table.
  *
- * Dump the entries in the reader lock table.
+ * [in] env     An environment handle returned by mdbx_env_create()
+ * [in] func    A MDBX_reader_list_func function
+ * [in] ctx     An arbitrary context pointer for the enumeration function.
  *
- * [in] env   An environment handle returned by mdbx_env_create()
+ * Returns A non-zero error value on failure and 0 on success,
- * [in] func  A MDBX_msg_func function
+ * or MDBX_RESULT_TRUE (-1) if the reader lock table is empty. */
- * [in] ctx   Anything the message function needs
+LIBMDBX_API int mdbx_reader_list(MDBX_env *env, MDBX_reader_list_func *func,
- *
+                                 void *ctx);
- * Returns < 0 on failure, >= 0 on success. */
-LIBMDBX_API int mdbx_reader_list(MDBX_env *env, MDBX_msg_func *func, void *ctx);
 
 /* Check for stale entries in the reader lock table.
  *
  * [in] env     An environment handle returned by mdbx_env_create()
  * [out] dead   Number of stale slots that were cleared
  *
- * Returns 0 on success, non-zero on failure. */
+ * Returns A non-zero error value on failure and 0 on success,
+ * or MDBX_RESULT_TRUE (-1) if a dead reader(s) found or mutex was recovered. */
 LIBMDBX_API int mdbx_reader_check(MDBX_env *env, int *dead);
 
 /* Returns a lag of the reading for the given transaction.

src/elements/core.c

@@ -13195,46 +13195,66 @@ int mdbx_set_dupsort(MDBX_txn *txn, MDBX_dbi dbi, MDBX_cmp_func *cmp) {
   return MDBX_SUCCESS;
 }
 
-int __cold mdbx_reader_list(MDBX_env *env, MDBX_msg_func *func, void *ctx) {
+int __cold mdbx_reader_list(MDBX_env *env, MDBX_reader_list_func *func,
-  char buf[64];
+                            void *ctx) {
-  int rc = 0, first = 1;
-
   if (unlikely(!env || !func))
-    return (MDBX_EINVAL > 0) ? -MDBX_EINVAL : MDBX_EINVAL;
+    return MDBX_EINVAL;
 
   if (unlikely(env->me_signature != MDBX_ME_SIGNATURE))
     return MDBX_EBADSIGN;
 
-  const MDBX_lockinfo *const lck = env->me_lck;
+  int rc = MDBX_RESULT_TRUE;
-  if (likely(lck)) {
+  int serial = 0;
-    const unsigned snap_nreaders = lck->mti_numreaders;
+  if (likely(env->me_lck)) {
+    const unsigned snap_nreaders = env->me_lck->mti_numreaders;
     for (unsigned i = 0; i < snap_nreaders; i++) {
-      if (lck->mti_readers[i].mr_pid) {
+      const MDBX_reader *r = env->me_lck->mti_readers + i;
-        const txnid_t txnid = lck->mti_readers[i].mr_txnid;
+    retry_reader:;
-        if (txnid == ~(txnid_t)0)
+      const mdbx_pid_t pid = r->mr_pid;
-          snprintf(buf, sizeof(buf), "%10" PRIuPTR " %" PRIxPTR " -\n",
+      if (!pid)
-                   (uintptr_t)lck->mti_readers[i].mr_pid,
+        continue;
-                   (uintptr_t)lck->mti_readers[i].mr_tid);
+      txnid_t txnid = r->mr_txnid;
-        else
+      const mdbx_tid_t tid = r->mr_tid;
-          snprintf(buf, sizeof(buf),
+      const pgno_t pages_used = r->mr_snapshot_pages_used;
-                   "%10" PRIuPTR " %" PRIxPTR " %" PRIaTXN "\n",
+      const uint64_t reader_pages_retired = r->mr_snapshot_pages_retired;
-                   (uintptr_t)lck->mti_readers[i].mr_pid,
+      mdbx_compiler_barrier();
-                   (uintptr_t)lck->mti_readers[i].mr_tid, txnid);
+      if (unlikely(pid != r->mr_pid || txnid != r->mr_txnid ||
+                   tid != r->mr_tid ||
+                   pages_used != r->mr_snapshot_pages_used ||
+                   reader_pages_retired != r->mr_snapshot_pages_retired))
+        goto retry_reader;
 
-        if (first) {
+      mdbx_assert(env, txnid > 0);
-          first = 0;
+      if (txnid == ~(txnid_t)0)
-          rc = func("    pid     thread     txnid\n", ctx);
+        txnid = 0;
-          if (rc < 0)
+
-            break;
+      size_t bytes_used = 0;
-        }
+      size_t bytes_retained = 0;
-        rc = func(buf, ctx);
+      uint64_t lag = 0;
-        if (rc < 0)
+      if (txnid) {
-          break;
+      retry_header:;
+        const MDBX_meta *const recent_meta = mdbx_meta_head(env);
+        const uint64_t head_pages_retired = recent_meta->mm_pages_retired;
+        const txnid_t head_txnid = mdbx_meta_txnid_fluid(env, recent_meta);
+        mdbx_compiler_barrier();
+        if (unlikely(recent_meta != mdbx_meta_head(env) ||
+                     head_pages_retired != recent_meta->mm_pages_retired) ||
+            head_txnid != mdbx_meta_txnid_fluid(env, recent_meta))
+          goto retry_header;
+
+        lag = head_txnid - txnid;
+        bytes_used = pgno2bytes(env, pages_used);
+        bytes_retained = (head_pages_retired > reader_pages_retired)
+                             ? pgno2bytes(env, (pgno_t)(head_pages_retired -
+                                                        reader_pages_retired))
+                             : 0;
       }
+      rc = func(ctx, ++serial, i, pid, tid, txnid, lag, bytes_used,
+                bytes_retained);
+      if (unlikely(rc != MDBX_SUCCESS))
+        break;
     }
   }
-  if (first)
-    rc = func("(no active readers)\n", ctx);
 
   return rc;
 }

src/tools/mdbx_stat.c

@@ -59,6 +59,27 @@ static void usage(char *prog) {
   exit(EXIT_FAILURE);
 }
 
+static int reader_list_func(void *ctx, int num, int slot, mdbx_pid_t pid,
+                            mdbx_tid_t thread, uint64_t txnid, uint64_t lag,
+                            size_t bytes_used, size_t bytes_retired) {
+  (void)ctx;
+  if (num == 1)
+    printf("Reader Table Status\n"
+           "   #\tslot\t%6s %*s %20s %10s %13s %13s\n",
+           "pid", (int)sizeof(size_t) * 2, "thread", "txnid", "lag", "used",
+           "retained");
+
+  printf(" %3d)\t[%d]\t%6" PRIdSIZE " %*" PRIxSIZE, num, slot, (size_t)pid,
+         (int)sizeof(size_t) * 2, (size_t)thread);
+  if (txnid)
+    printf(" %20" PRIu64 " %10" PRIu64 " %12.1fM %12.1fM\n", txnid, lag,
+           bytes_used / 1048576.0, bytes_retired / 1048576.0);
+  else
+    printf(" %20s %10s %13s %13s\n", "-", "0", "0", "0");
+
+  return user_break ? MDBX_RESULT_TRUE : MDBX_RESULT_FALSE;
+}
+
 int main(int argc, char *argv[]) {
   int o, rc;
   MDBX_env *env;
@@ -206,13 +227,24 @@ int main(int argc, char *argv[]) {
   }
 
   if (rdrinfo) {
-    printf("Reader Table Status\n");
+    rc = mdbx_reader_list(env, reader_list_func, nullptr);
-    rc = mdbx_reader_list(env, (MDBX_msg_func *)fputs, stdout);
+    if (rc == MDBX_RESULT_TRUE)
-    if (rdrinfo > 1) {
+      printf("Reader Table is empty\n");
+    else if (rc == MDBX_SUCCESS && rdrinfo > 1) {
       int dead;
-      mdbx_reader_check(env, &dead);
+      rc = mdbx_reader_check(env, &dead);
-      printf("  %d stale readers cleared.\n", dead);
+      if (rc == MDBX_RESULT_TRUE) {
-      rc = mdbx_reader_list(env, (MDBX_msg_func *)fputs, stdout);
+        printf("  %d stale readers cleared.\n", dead);
+        rc = mdbx_reader_list(env, reader_list_func, nullptr);
+        if (rc == MDBX_RESULT_TRUE)
+          printf("  Now Reader Table is empty\n");
+      } else
+        printf("  No stale readers.\n");
+    }
+    if (MDBX_IS_ERROR(rc)) {
+      fprintf(stderr, "mdbx_txn_begin failed, error %d %s\n", rc,
+              mdbx_strerror(rc));
+      goto env_close;
     }
     if (!(subname || alldbs || freinfo))
       goto env_close;