Hi David,
Thank you for your comments.
On 2019/10/26 0:36, David Marchand wrote:
On Fri, Oct 11, 2019 at 11:36 AM David Marchand
<david.march...@redhat.com> wrote:
Some comments.
ping.
The title does not reflect the observed issue.
I would like to consider to revise it.
I understand that secondary processeses can't be started from a docker
container.
I've confirmed that secondary process can be started as a container app
with SPP, and DPDK v18.08 and v19.08. SPP is a multi-process app
supporting container usecases.
http://git.dpdk.org/apps/spp/
The patch title should reflect this.
On Wed, Jul 24, 2019 at 10:20 AM <yasufu...@gmail.com> wrote:
From: Yasufumi Ogawa <ogawa.yasuf...@lab.ntt.co.jp>
In secondary_msl_create_walk(), it creates a file for fbarrays with its
PID for reserving unique name among secondary processes. However, it
does not work if secondary is run as app container because each of
containerized secondary has PID 1. To reserve unique name, use hostname
instead of PID because hostname is assigned as a short form of 64
digits full container ID in docker.
Cc: sta...@dpdk.org
I don't think we want to backport this behavior change.
This issue was included from DPDK v18.08, and some users of SPP are
still using stable 18.11. So, I would appreciate if you agree to backport.
Signed-off-by: Yasufumi Ogawa <ogawa.yasuf...@lab.ntt.co.jp>
---
lib/librte_eal/linux/eal/eal_memalloc.c | 28 +++++++++++++++++++++++--
1 file changed, 26 insertions(+), 2 deletions(-)
diff --git a/lib/librte_eal/linux/eal/eal_memalloc.c
b/lib/librte_eal/linux/eal/eal_memalloc.c
index 1f6a7c18f..356b304a8 100644
--- a/lib/librte_eal/linux/eal/eal_memalloc.c
+++ b/lib/librte_eal/linux/eal/eal_memalloc.c
@@ -1366,6 +1366,7 @@ secondary_msl_create_walk(const struct rte_memseg_list
*msl,
struct rte_memseg_list *primary_msl, *local_msl;
char name[PATH_MAX];
int msl_idx, ret;
+ char proc_id[33] = { 0 }; /* 32bytes is enough if using hostname */
This variable only makes sense in the if (getpid() == 1) branch,
please move it there, and see below comment about using gethostname().
Sure. It works correctly in secondary app container and I should replace it.
if (msl->external)
return 0;
@@ -1375,8 +1376,31 @@ secondary_msl_create_walk(const struct rte_memseg_list
*msl,
local_msl = &local_memsegs[msl_idx];
/* create distinct fbarrays for each secondary */
- snprintf(name, RTE_FBARRAY_NAME_LEN, "%s_%i",
- primary_msl->memseg_arr.name, getpid());
+ /* If run secondary in a container, the name of fbarray file should
+ * not be decided with pid because getpid() always returns 1.
+ * In docker, hostname is assigned as a short form of full container
+ * ID. So use hostname as unique ID among containers instead.
I understand this is how it works for docker.
Is this the same in other container environments?
Umm... I've tested on other than docker because I don't have test
environment. I am not sure which of container runtimes should be
supported actually. I think it is enough as the first step to fix this
issue of docker. Moreover, the essential problem is that getpid()
returns 1 in docker or not.
I am also not sure which of environments other than docker should be
supported if necessary. What do you think?
+ */
+ if (getpid() == 1) {
+ FILE *hn_fp;
+ hn_fp = fopen("/etc/hostname", "r");
Why not use gethostname() ?
Plus, this api defines the maximum size of the hostname as HOST_NAME_MAX bytes.
Yes. I should use gethostname() and replace hardcoded size with
HOST_NAME_MAX.
Thanks,
Yasufumi
+ if (hn_fp == NULL) {
+ RTE_LOG(ERR, EAL,
+ "Cannot open '/etc/hostname' for secondary\n");
+ return -1;
+ }
+
+ /* with docker, /etc/hostname just has one entry of hostname */
+ if (fscanf(hn_fp, "%32s", proc_id) == EOF) {
+ fclose(hn_fp);
+ return -1;
+ }
+ fclose(hn_fp);
+ } else
+ sprintf(proc_id, "%d", (int)getpid());
+
+ snprintf(name, RTE_FBARRAY_NAME_LEN, "%s_%s",
+ primary_msl->memseg_arr.name, proc_id);
ret = rte_fbarray_init(&local_msl->memseg_arr, name,
primary_msl->memseg_arr.len,
