rbd: fix rbd children listing when child is in trash and add rbd_list_children2 method #18483

Songweibin · 2017-10-23T12:54:16Z

Add rbd_list_children2 and librbd::RBD::list_children2 methods as @dillaman suggested in #14865 (comment)
and after this fix rbd children listing as follows:

root@~:/ceph-dev/build# ./bin/rbd children i1@s1 
rbd/i3
root@~:/ceph-dev/build# ./bin/rbd children i1@s1 -a
rbd/i2 (trash 103c3d1b58ba)
rbd/i3
root@~:/ceph-dev/build#

http://tracker.ceph.com/issues/21893
Signed-off-by: songweibin song.weibin@zte.com.cn

xiexingguo · 2017-10-24T00:42:17Z

need to fix the cli test first..

Songweibin · 2017-10-25T09:52:40Z

@dillaman please help to have a review, thanks in advance.

dillaman

Thanks -- can you also ensure the test cases verify that images in the trash are properly returned?

dillaman · 2017-10-25T13:11:00Z

src/include/rbd/librbd.hpp

+    std::string image_name;
+    std::string image_id;
+  } children_info_t;
+


Nit: child_info_t (since children implies more than one)

dillaman · 2017-10-25T13:11:34Z

src/include/rbd/librbd.hpp

+   * Returns a structure of poolname, imagename, imageid for 
+   * each clone of this image at the currently set snapshot.
+   */
+  int list_children2(std::vector<librbd::children_info_t>& children2);


Nit: no need for 2 at the end of the parameter name

dillaman · 2017-10-25T13:12:52Z

src/librbd/internal.h

@@ -105,6 +105,8 @@ namespace librbd {
  int list(librados::IoCtx& io_ctx, std::vector<std::string>& names);
  int list_children(ImageCtx *ictx,
 		    std::set<std::pair<std::string, std::string> > & names);
+  int list_children2(ImageCtx *ictx,


Let's not add a duplicate function within the internal API -- instead, update the existing list_children method to take std::vector<child_info_t> and update the public API methods to convert between the new type and the old pair.

dillaman · 2017-10-25T13:13:13Z

src/librbd/librbd.cc

@@ -1272,6 +1272,20 @@ namespace librbd {
    return r;
  }

+  int Image::list_children2(vector<librbd::children_info_t>& children2)


Nit: no need for 2 suffix on param name

dillaman · 2017-10-25T13:17:35Z

src/librbd/internal.cc

+          lderr(cct) << "Error looking up name for image id " << id_it
+                     << " in pool " << info.first.second << dendl;
+          return r;
+        } else if (r == -ENOENT) {


Nit: execute the -ENOENT handling first. If it fails w/ -EOENT, you should allow it to fall through to the generic handling so that the error message is "error looking up name for image id XYZ in pool" instead of "in rbd trash".

dillaman · 2017-10-25T13:18:43Z

src/librbd/internal.cc

+        } else if (r == -ENOENT) {
+          cls::rbd::TrashImageSpec trash_spec;
+          r = cls_client::trash_get(&ioctx, id_it, &trash_spec);
+          if (r < 0) {


Need to support older OSDs to filter out -EOPNOTSUPP errors and change them to -ENOENT so that it falls down into the moved generic error handler clause (from above).

dillaman · 2017-10-25T13:23:35Z

src/tools/rbd/action/Children.cc

-      f->close_section();
-    } else {
-      std::cout << child_it.first << "/" << child_it.second << std::endl;
+  if (all_flag) {


Nit: instead of duplicating all this logic, perhaps always use the new list_children2 API and if --all wasn't specified, don't print any children that are in the trash.

dillaman · 2017-10-25T13:25:24Z

src/include/rbd/librbd.hpp

+    std::string pool_name;
+    std::string image_name;
+    std::string image_id;
+  } children_info_t;


Perhaps add a bool trash member so that users don't need to re-execute a trash_get (since it was already performed within the API).

dillaman · 2017-10-25T13:26:15Z

src/pybind/rbd/rbd.pyx

@@ -2232,6 +2235,40 @@ written." % (self.name, ret, length))
            free(c_pools)
            free(c_images)

+    def list_children2(self):


Might be nicer to switch to an iterator style and return a dict for each child.

Songweibin · 2017-10-26T09:20:33Z

@dillaman Thanks for your nicer suggestions, will update later.

Songweibin · 2017-10-30T05:37:27Z

@dillaman Updated, please take a look, thanks.
But I'm not sure I follow your idea about this:

Let's not add a duplicate function within the internal API -- instead, update the existing list_children method to take std::vector<child_info_t> and update the public API methods to convert between the new type and the old pair.

dillaman

@Songweibin The function list_children(ImageCtx *ictx, set<pair<string, string> >& names) should be removed from internal.h/cc and all "list children" public API methods should use list_children(ImageCtx *ictx, vector<child_info_t>& names)

dillaman · 2017-10-30T15:22:52Z

src/librbd/internal.cc

+          lderr(cct) << "Error looking up name for image id " << id_it
+                     << " in pool " << info.first.second << dendl;
+          return r;
+        } else if (r == -ENOENT) {


Comment still needs to be addressed

dillaman · 2017-10-30T15:24:37Z

src/include/rbd/librbd.h

+CEPH_RBD_API int rbd_list_children2(rbd_image_t image,
+                                    rbd_child_info_t *children,
+                                    int *max_children);
+CEPH_RBD_API void rbd_list_child_end(rbd_child_info_t *children);


Nit: rename to rbd_list_child_cleanup and add a size_t num_children parameter so that all the initialized rbd_child_info_t structs can be cleaned

dillaman · 2017-10-30T15:28:49Z

src/librbd/internal.cc

 	names.insert(make_pair(info.first.second, name));
      }
    }

    return 0;
  }

+  bool compare_by_name(const child_info_t& c1, const child_info_t& c2) {


Nit: add to anonymous namespace near top of file

dillaman · 2017-10-30T15:30:55Z

src/librbd/internal.h

@@ -105,6 +105,9 @@ namespace librbd {
  int list(librados::IoCtx& io_ctx, std::vector<std::string>& names);
  int list_children(ImageCtx *ictx,
 		    std::set<std::pair<std::string, std::string> > & names);
+  int list_children(ImageCtx *ictx,
+                    std::vector<child_info_t>& names);


Pass out-values by pointer instead of reference

Signed-off-by: songweibin <song.weibin@zte.com.cn>

Songweibin · 2017-10-31T12:35:21Z

@dillaman Updated all, please take a review again, thanks.

dillaman · 2017-11-02T02:48:51Z

src/include/rbd/librbd.hpp

   */
-  int list_children(std::set<std::pair<std::string, std::string> > *children);
+  int list_children(std::vector<librbd::child_info_t> *children);


You cannot change the public API -- you can only change the private API that is implemented in "src/librbd/internal.cc"

Oh, sorry, at first I thought list_children(std::set<std::pair<std::string, std::string> > *children) was unused, so i changed it directly. Actually there may be other external APPs used it. Already modified, thanks.

dillaman · 2017-11-02T02:51:10Z

src/test/cli/rbd/help.t

@@ -144,7 +144,7 @@ Skip test on FreeBSD as it generates different output there.
    --io-type arg        IO type (read , write, or readwrite(rw))

  rbd help children
-  usage: rbd children [--pool <pool>] [--image <image>] [--snap <snap>] 
+  usage: rbd children [--pool <pool>] [--image <image>] [--snap <snap>] [--all] 


These (related) changes seem to be mixing and matching between commits. The title of this commit is to fix a bug re: listing children for trashed images so it should be limited to handling that fix. The next commit can add the new API methods and expand the rbd CLI to list trashed images.

Signed-off-by: songweibin <song.weibin@zte.com.cn>

Songweibin · 2017-11-02T06:33:37Z

@dillaman Done, please help to take a look, thank you.

dillaman · 2017-11-06T14:54:05Z

src/librbd/librbd.cc

+    children[i].trash = cpp_children[i].trash;
+    if (!children[i].image_id) {
+      for (int j = 0; j < i; j++)
+      free((void *)children[j].image_id);


Nit: re-use rbd_list_children_cleanup

@dillaman fixed, thanks for your carefully review.

dillaman · 2017-11-06T14:54:50Z

src/librbd/librbd.cc

+    children[i].image_name = strdup(cpp_children[i].image_name.c_str());
+    children[i].image_id = strdup(cpp_children[i].image_id.c_str());
+    children[i].trash = cpp_children[i].trash;
+    if (!children[i].image_id) {


Nit: why only validate the result of image_id instead of also checking pool_name and image_name?

dillaman · 2017-11-07T15:14:39Z

src/librbd/librbd.cc

+  int r = librbd::list_children(ictx, &cpp_children);
+  if (r == -ENOENT) {
+    tracepoint(librbd, list_children_exit, *max_children);
+    return 0;


Nit: if r == -NOENT, just set r = 0 and continue so that max_return is properly populated and the array is properly null terminated.

dillaman · 2017-11-07T15:15:01Z

src/librbd/librbd.cc

+    children[i].image_name = strdup(cpp_children[i].image_name.c_str());
+    children[i].image_id = strdup(cpp_children[i].image_id.c_str());
+    children[i].trash = cpp_children[i].trash;
+    if (!children[i].pool_name && !children[i].image_name &&


Swap the && operators to || since it should fail if any memory allocation fails

@dillaman fixed, thanks.

dillaman

lgtm

dillaman · 2017-11-08T15:48:22Z

src/tools/rbd/action/Children.cc

-      std::cout << child_it.first << "/" << child_it.second << std::endl;
+      if (all_flag) {
+        std::cout << it->pool_name << "/" << it->image_name << "/"
+                  << it->image_id;


Sorry -- I missed this during the review but saw it when play-testing the change. I don't like how adding '--all" changes the format to include the image id. The "/" is part of the image-spec (i.e. pool/image) but it's being overloaded here to include the image id. I would include the image id only for items in the trash like the following:

<pool name>/<image name> (trash <image id>)

Yeah, sounds more reasonable. Already fixed, thanks.

dillaman · 2017-11-13T19:00:14Z

The python tests appear to be failing:

http://qa-proxy.ceph.com/teuthology/jdillaman-2017-11-13_12:10:01-rbd-wip-jd-testing2-distro-basic-smithi/1844437/teuthology.log

…ages implement librbd::RBD::list_children2 and rbd_list_children2 methods and expand the rbd CLI to list trashed images Signed-off-by: songweibin <song.weibin@zte.com.cn>

Songweibin · 2017-11-14T10:40:59Z

@dillaman Sorry, since list_children2() returns an iterator style it shoud convert to list when comparing, and there also have a mistake when manipulating expected_children/expected_children2 in the trash cases.
All have fixed, thanks.

Songweibin force-pushed the wip-list-children2 branch from 1e74946 to 6332433 Compare October 24, 2017 02:11

xiexingguo requested a review from dillaman October 24, 2017 10:15

xiexingguo added feature rbd labels Oct 24, 2017

dillaman reviewed Oct 25, 2017

View reviewed changes

Songweibin force-pushed the wip-list-children2 branch 7 times, most recently from e8399a8 to f541e87 Compare October 30, 2017 04:39

dillaman reviewed Oct 30, 2017

View reviewed changes

Songweibin force-pushed the wip-list-children2 branch 2 times, most recently from 88d6557 to 0d2c611 Compare October 31, 2017 10:24

librbd: drop unused 'rbd_snap_t'

d929e60

Signed-off-by: songweibin <song.weibin@zte.com.cn>

Songweibin changed the title ~~rbd: add librbd::RBD::list_children2 and rbd_list_children2 methods~~ rbd: fix rbd children listing when child is in trash and add rbd_list_children2 method Oct 31, 2017

Songweibin force-pushed the wip-list-children2 branch from 0d2c611 to 007d4d9 Compare October 31, 2017 12:30

dillaman reviewed Nov 2, 2017

View reviewed changes

Songweibin force-pushed the wip-list-children2 branch from 007d4d9 to 1d603b2 Compare November 2, 2017 05:06

librbd: fix rbd children listing when child is in trash bin

4803069

Signed-off-by: songweibin <song.weibin@zte.com.cn>

Songweibin force-pushed the wip-list-children2 branch from 1d603b2 to e6332d0 Compare November 2, 2017 05:15

dillaman reviewed Nov 6, 2017

View reviewed changes

Songweibin force-pushed the wip-list-children2 branch 2 times, most recently from 59a20ad to 5856920 Compare November 7, 2017 05:35

dillaman reviewed Nov 7, 2017

View reviewed changes

Songweibin force-pushed the wip-list-children2 branch from 5856920 to e4ea1f1 Compare November 8, 2017 00:38

dillaman approved these changes Nov 8, 2017

View reviewed changes

dillaman added needs-qa and removed needs-qa labels Nov 8, 2017

dillaman reviewed Nov 8, 2017

View reviewed changes

Songweibin force-pushed the wip-list-children2 branch from e4ea1f1 to ed3753c Compare November 9, 2017 09:05

dillaman approved these changes Nov 9, 2017

View reviewed changes

dillaman added needs-qa wip-jason-testing and removed needs-qa wip-jason-testing labels Nov 9, 2017

librbd: add new API methods and expand the rbd CLI to list trashed im…

9d9a276

…ages implement librbd::RBD::list_children2 and rbd_list_children2 methods and expand the rbd CLI to list trashed images Signed-off-by: songweibin <song.weibin@zte.com.cn>

Songweibin force-pushed the wip-list-children2 branch from ed3753c to 9d9a276 Compare November 14, 2017 06:17

dillaman added the wip-jason-testing label Nov 14, 2017

dillaman merged commit 16715ec into ceph:master Nov 14, 2017

Songweibin deleted the wip-list-children2 branch November 15, 2017 00:56

rbd: fix rbd children listing when child is in trash and add rbd_list_children2 method #18483

rbd: fix rbd children listing when child is in trash and add rbd_list_children2 method #18483

Conversation

Songweibin commented Oct 23, 2017 • edited

xiexingguo commented Oct 24, 2017

Songweibin commented Oct 25, 2017

dillaman left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Songweibin commented Oct 26, 2017

Songweibin commented Oct 30, 2017

dillaman left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Songweibin commented Oct 31, 2017

Choose a reason for hiding this comment

Songweibin Nov 2, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Songweibin commented Nov 2, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dillaman left a comment

Choose a reason for hiding this comment

dillaman Nov 8, 2017 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dillaman commented Nov 13, 2017

Songweibin commented Nov 14, 2017 • edited

Songweibin commented Oct 23, 2017 •

edited

Songweibin Nov 2, 2017 •

edited

dillaman Nov 8, 2017 •

edited

Songweibin commented Nov 14, 2017 •

edited