Changeset 33365

Timestamp:

Feb 24, 2012, 10:04:28 AM (14 years ago)

Author:

eugene

Message:

use a remote md5sum call via pclient to check on successful transfer (do not trust NFS...)

Location:

branches/eam_branches/ipp-20111122/Ohana/src/photdbc

Files:

• Makefile (modified) (1 diff)
• include/dvodist.h (modified) (2 diffs)
• src/CopyToHostLocation.c (modified) (4 diffs)
• src/HostTableLoad.c (modified) (3 diffs)
• src/PclientCommands.c (added)

branches/eam_branches/ipp-20111122/Ohana/src/photdbc/Makefile

@@ -42 +42 @@
 $(SRC)/AssignSkyToHost.$(ARCH).o        \
 $(SRC)/CopyToHostLocation.$(ARCH).o     \
-$(SRC)/CopyFromHostLocation.$(ARCH).o
+$(SRC)/CopyFromHostLocation.$(ARCH).o   \
+$(SRC)/PclientCommands.$(ARCH).o
 
 OLD = \

branches/eam_branches/ipp-20111122/Ohana/src/photdbc/include/dvodist.h

@@ -4 +4 @@
 # include <md5.h>
 # include <unistd.h> // needed?
+
+# define PCLIENT_PROMPT "pclient:"
 
 // used by the md5 stuff
@@ -37 +39 @@
 
 int           get_md5_with_copy (char *input, char *output, md5_byte_t *digest);
+
+int CheckBusyJob (HostInfo *host, IOBuffer *stdout_buf, IOBuffer *stderr_buf);
+int GetJobOutput (char *command, HostInfo *host, IOBuffer *output, int size);
+int PclientResponse (HostInfo *host, char *response, IOBuffer *buffer);
+int PclientCommand (HostInfo *host, char *command, IOBuffer *buffer);

branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/CopyToHostLocation.c

@@ -4 +4 @@
 # define DELETE_ORIGINAL 0
 // make this a user option
+
+int get_md5_from_remote (HostInfo *host, char *filename, md5_byte_t *digest);
+int get_md5_from_pclient (HostInfo *host, char *filename, md5_byte_t *digest);
+int hexchar_to_int (char input);
 
 // XXX collapse this and CopyFromHostLocation (only difference is name of src/tgt and bits set on success)
@@ -95 +99 @@
         
       // need to re-open and re-read tgtname to check md5sum
-      // XXX this would be safer and faster if we used the remote shell to perform the md5sum locally.  
-      if (!get_md5_with_copy (tgtname, NULL, tgtDigest)) {
+      char *absname = abspath (tgtname, 1024);
+      if (!get_md5_from_pclient (host, absname, tgtDigest)) {
         fprintf (stderr, "error reading %s or getting md5\n", tgtname);
         success = FALSE;
@@ -135 +139 @@
         sprintf (tgtname, "%s/%s.%s", host->pathname, skylist->regions[i]->name, extname[j]);
         sprintf (srcname, "%s/%s.%s", catdir, skylist->regions[i]->name, extname[j]);
-
-        // do not rename or unlink original if it does not exist..
-        status = stat (srcname, &filestat);
-        if (status && (errno == ENOENT)) continue;  
-        
+ 
+        // do not rename or unlink original if it does not exist..
+        status = stat (srcname, &filestat);
+        if (status && (errno == ENOENT)) continue;  
+
         if (!DEBUG && DELETE_ORIGINAL) {
           unlink (srcname);
@@ -190 +194 @@
  * read the SPLIT headers to see if MEASURE, MISSING, SECFILT fields exist?
  */
+
+// XXX this would be better controlled if we used a remote pclient.  then we could
+// check things like the exit status and 
+int get_md5_from_remote (HostInfo *host, char *filename, md5_byte_t *digest) {
+
+  IOBuffer buffer;
+
+  char line[1024];
+  int Nline = snprintf (line, 1024, "md5sum %s\n", filename);
+  assert (Nline < 1024);
+
+  write (host->stdio[0], line, Nline);
+
+  InitIOBuffer (&buffer, 100);
+  int status = EmptyIOBuffer (&buffer, 10000, host->stdio[1]);
+  fprintf (stderr, "md5sum status: %d\n", status);
+      
+  fprintf (stderr, "buffer: %s\n", buffer.buffer);
+
+  char result[1024], fileout[1024];
+  int Nscan = sscanf (buffer.buffer, "%s %s", result, fileout);
+  assert (Nscan == 2);
+  assert (strlen(result) == NDIGEST);
+  
+  memcpy (digest, result, NDIGEST);
+  return TRUE;
+}
+
+// pclient version of remote client
+int get_md5_from_pclient (HostInfo *host, char *filename, md5_byte_t *digest) {
+
+  IOBuffer buffer, stdout_buf, stderr_buf;
+
+  char line[1024];
+  int Nline = snprintf (line, 1024, "job md5sum %s\n", filename);
+  assert (Nline < 1024);
+
+  // fprintf (stderr, "command: %s\n", line);
+
+  // start the md5sum command
+  InitIOBuffer (&buffer, 100);
+  PclientCommand (host, line, &buffer);
+  PclientResponse (host, PCLIENT_PROMPT, &buffer);
+  // XXX for the moment, abort if anything goes wrong
+
+  // wait for result
+  while (!CheckBusyJob (host, &stdout_buf, &stderr_buf)) {
+    usleep (10000);
+  }
+  if (stderr_buf.Nbuffer) {
+    fprintf (stderr, "error in md5sum: %s\n", stderr_buf.buffer);
+    exit (1);
+  }
+
+  // fprintf (stderr, "result from md5sum: %s\n", stdout_buf.buffer);
+
+  char result[1024], fileout[1024];
+  int Nscan = sscanf (stdout_buf.buffer, "%s %s", result, fileout);
+  assert (Nscan == 2);
+  assert (strlen(result) == 2*NDIGEST);
+  
+  // result is the set of 0-f nibbles, convert to the digest
+  int i;
+  for (i = 0; i < NDIGEST; i++) {
+    int value1 = hexchar_to_int (result[2*i+0]);
+    int value2 = hexchar_to_int (result[2*i+1]);
+    digest[i] = (value1 << 4) + value2;
+    // fprintf (stderr, "%02x %c %c\n", digest[i], result[2*i], result[2*i+1]);
+  }
+  
+  // need to reset pclient for next command...
+  PclientCommand (host, "reset", &buffer);
+  PclientResponse (host, PCLIENT_PROMPT, &buffer);
+
+  return TRUE;
+}
+
+int hexchar_to_int (char input) {
+
+  switch (input) {
+    case '0':
+    case '1':
+    case '2':
+    case '3':
+    case '4':
+    case '5':
+    case '6':
+    case '7':
+    case '8':
+    case '9':
+      return input - '0';
+    case 'a':
+    case 'b':
+    case 'c':
+    case 'd':
+    case 'e':
+    case 'f':
+      return input - 'a' + 10;
+    case 'A':
+    case 'B':
+    case 'C':
+    case 'D':
+    case 'E':
+    case 'F':
+      return input - 'A' + 10;
+    default:
+      abort();
+  }
+}

branches/eam_branches/ipp-20111122/Ohana/src/photdbc/src/HostTableLoad.c

@@ -5 +5 @@
 
   int i;
-  int stdio[3];
   char command[64];
   char shell[64];
 
   strcpy (command, "ssh");
-  strcpy (shell, "csh");
+  strcpy (shell, "pclient");
 
   for (i = 0; i < table->Nhosts; i++) {
@@ -18 +17 @@
 
     int errorInfo;
-    int pid = rconnect (command, table->hosts[i].hostname, shell, stdio, &errorInfo, TRUE);
+    int pid = rconnect (command, table->hosts[i].hostname, shell, table->hosts[i].stdio, &errorInfo, TRUE);
     if (!pid) {     
       /** failure to start: extend retry period **/
@@ -24 +23 @@
       exit (1);
     }
-
-    // XXX check anything else?
-
-    // close the connection
-    close(stdio[0]);
-    close(stdio[1]);
-    close(stdio[2]);
+    table->hosts[i].pid = pid;
+    // keep the connection open to use for md5sum check
 
     int status = check_dir_access (table->hosts[i].pathname, DEBUG);