Add pthreads xfer, axl_cp, and update AXL_Add for directories

[tonyhutter]

• Add a new AXL_XFER_PTHREAD transfer. This uses multiple threads to transfer the files.

• Add an axl_cp command. axl_cp works like the standard cp command, but allows you to specify the AXL transfer type (sync, pthreads, bbapi, etc...). It will also serve as good example code for AXL users. Along with this, test_axl.c has been replaced with a test_axl bash script. This will make it a lot easier to write test cases. The new test_axl script will test both sync and pthread transfers via calls to axl_cp.

• Allow passing directories to AXL_Add(). If a directory is passed, all files and subdirs within the dir are recursively copied. Basically, AXL_Add() will follow the same semantics as the cp command. This is conceptually easier for the user to understand, and saves them from having to traverse their directories manually. Note that the recursive directory copy will currently not copy empty subdirectories.

[tonyhutter]

Travis caught this: ../src/libaxl.so: undefined reference to 'pthread_join' Looks like I forgot to update CMakeLists to include -lpthread. Let me fix that...

[tonyhutter]

Travis is hitting a make check failure when running axl_cp. I'll need to update that too

[tonyhutter]

I needed a make check test case for Travis to run, so I wrote a test_axl script which invokes axl_cp. I've included the test_axl script in my latest push. I've also updated my PR comments to reference it.

make check works in my Fedora 29 VM, but it's failing on Travis, so I need to dig into that.

[tonyhutter]

Looks like Travis is erroring out here:

1: Test command: /home/travis/build/ECP-VeloC/AXL/test/test_axl
1: Test timeout computed to be: 9.99988e+06
1: Testing sync transfer... failed copy
1: AXL 0.3.0 ERROR: travis-job-316ad4c6-c516-4ece-a8ba-645769a85db5: axl_async_init_daemon: child process failed to launch: axld /dev/shm/axld: No such file or directory @ axl_async_init_daemon /home/travis/build/ECP-VeloC/AXL/src/axl_async_daemon.c:578
1: AXL_Init() failed (error -1)
1: AXL 0.3.0 ERROR: travis-job-316ad4c6-c516-4ece-a8ba-645769a85db5: axl_async_init_daemon: child process failed to start: axld /dev/shm/axld @ axl_async_init_daemon /home/travis/build/ECP-VeloC/AXL/src/axl_async_daemon.c:601
1: AXL_Init() failed (error -1)
1/1 Test #1: transfer_tests ...................***Failed    5.67 sec

I can reproduce this on my VM when I build with -DAXL_ASYNC_DAEMON=ON like Travis does. It appears to be an existing bug that's unrelated to my change, as I get the same error when running test_axl -s in the master branch:

$ ./test/test_axl -s
Using existing source file /tmp/testfile
AXL 0.3.0 ERROR: fedora29: axl_async_init_daemon: child process failed to launch: axld /dev/shm/axld: No such file or directory @ axl_async_init_daemon /home/hutter/AXL-master/src/axl_async_daemon.c:578
AXL 0.3.0: fedora29: axl_sync_start: Read and copied test_file to /home/hutter/AXL-master/build/test_file_moved with success code 0 @ axl_sync_start /home/hutter/AXL-master/src/axl_sync.c:46

The only reason that Travis is passing on master is that make check is incorrectly calling test_axl with no arguments, and so no tests get run, and thus this axl_async_init_daemon() code is never invoked.

Since this is broken, I'm going to remove -DAXL_ASYNC_DAEMON=ON from .travis.yml.

[CamStan]

It's not building for me when I pull your branch. Tagging @gonsie to review the cmake.

[tonyhutter]

@CamStan can you pull again and try building? If it's still not working, can you post the build failure?

[CamStan]

It built for me this time. Might have been something on my end that caused my failure last week.

[tonyhutter]

@gonsie thanks AXL_EXTERNAL_LIBS did the trick. It's building now.

[gonsie]

@tonyhutter how do I submit changes back to your branch? I'm not familiar with the all-in-one-commit model.

[tonyhutter]

Hmm.. I don't know if you can push to my branch. You can post you patch directly in a comment and I can apply it. That way all the reviewers can see your proposed change. I've done that in the past for some ZFS code reviews.

[tonyhutter]

Alternatively, you can create a branch with your patch in it and point me to it.

[gonsie]

From 186e5e0dfed377d5d89733541c0fc13b1613d531 Mon Sep 17 00:00:00 2001
From: "Elsa Gonsiorowski (Uranus)" <gonsiorowski1@llnl.gov>
Date: Fri, 31 May 2019 10:11:43 -0700
Subject: [PATCH 1/2] get_nprocs on apple

---
 src/axl_pthread.c | 19 ++++++++++++++++++-
 1 file changed, 18 insertions(+), 1 deletion(-)

diff --git a/src/axl_pthread.c b/src/axl_pthread.c
index 3a2950b..d128f84 100644
--- a/src/axl_pthread.c
+++ b/src/axl_pthread.c
@@ -1,11 +1,18 @@
 #include <stdio.h>
 #include <stdlib.h>
 #include <pthread.h>
-#include <sys/sysinfo.h>    /* get_nprocs() */
 #include "axl_internal.h"
 #include "kvtree_util.h"
 #include "axl_sync.h"

+/* get_nprocs() */
+#if defined(__APPLE__)
+#include <sys/sysctl.h>
+#else
+#include <sys/sysinfo.h>
+#endif
+
+
 #define MIN(a,b) (a < b ? a : b)

 /*
@@ -200,7 +207,17 @@ int axl_pthread_start (int id)
     file_count = kvtree_size(files);

     /* Get number of hardware threads on our CPU (includes HT) */
+#if defined(__APPLE__)
+    int count;
+    size_t size = sizeof(count);
+    if (sysctlbyname("hw.ncpu",&count,&size,NULL,0)) {
+        cpu_threads = 1;
+    } else {
+        cpu_threads = count;
+    }
+#else
     cpu_threads = get_nprocs();
+#endif

     threads = MIN(cpu_threads, MIN(file_count, MAX_THREADS));

-- 
2.17.2 (Apple Git-113)

[gonsie]

Could you also change the test_axl script name to test_axl.sh?

[tonyhutter]

@gonsie thanks for the patch! I'll apply it and retest. I'll also do the test_axl.sh rename.

[gonsie]

The test is failing on my machine... the temp files that are created cannot be opened by the AXL library (even though they seem to exist). But, I'll stop fiddling with this PR and approve it. We can work on testing later.

[tonyhutter]

@gonsie I included your changes in my latest push.

[gonsie]

looks like the patch I submitted didn't apply cleanly and the sys/sysinfo.h include is still in there. FYI, I just pushed a fix to master.