| Message ID | 20210411114543.2003514-1-roid@nvidia.com |
|---|---|
| Headers | show |
| Series | reduce memory bloat and aging active ct conns | expand |
On 11.04.21 13:45, Roi Dayan wrote: > BugLink: https://bugs.launchpad.net/bugs/1922678 > > > SRU Justification: > > The patches are fixing a possible memory hog and aging active ct conns. > > * Explain the bug(s) > > Memory bloat and aging active ct conns can happen from any stress test > as we have a single workqueue for handling the ct offload conns > for add/del/stats actions. > > * brief explanation of fixes > > The fix is splitting add/del/stats for diff workqueue and also > we set a limit for add work entries. > > The first patch is already upstream. > > * How to test > > Testing was done with just stress traffic with lots of conns different 5-tuple. > > * What it could break. > > Can take a lot of memory for a long time and also causing active conns to > age out when not necessary. > > > Oz Shlomo (1): > netfilter: flowtable: separate replace, destroy and stats to different > workqueues > > Roi Dayan (1): > UBUNTU: SAUCE: netfilter: nf_flow_table_offload: Limit work entries on > offload add wq > > net/netfilter/nf_flow_table_offload.c | 56 +++++++++++++++++++++++---- > 1 file changed, 48 insertions(+), 8 deletions(-) > I did update the bug report for this set the way it would be expected (what Tim asked to be done for the other two). When doing so I did slightly adjust the section about the patches. Acked-by: Stefan Bader <stefan.bader@canonical.com>
Acked-by: Tim Gardner <tim.gardner@canonical.com> On 4/11/21 5:45 AM, Roi Dayan wrote: > BugLink: https://bugs.launchpad.net/bugs/1922678 > > > SRU Justification: > > The patches are fixing a possible memory hog and aging active ct conns. > > * Explain the bug(s) > > Memory bloat and aging active ct conns can happen from any stress test > as we have a single workqueue for handling the ct offload conns > for add/del/stats actions. > > * brief explanation of fixes > > The fix is splitting add/del/stats for diff workqueue and also > we set a limit for add work entries. > > The first patch is already upstream. > > * How to test > > Testing was done with just stress traffic with lots of conns different 5-tuple. > > * What it could break. > > Can take a lot of memory for a long time and also causing active conns to > age out when not necessary. > > > Oz Shlomo (1): > netfilter: flowtable: separate replace, destroy and stats to different > workqueues > > Roi Dayan (1): > UBUNTU: SAUCE: netfilter: nf_flow_table_offload: Limit work entries on > offload add wq > > net/netfilter/nf_flow_table_offload.c | 56 +++++++++++++++++++++++---- > 1 file changed, 48 insertions(+), 8 deletions(-) >
On 11.04.21 13:45, Roi Dayan wrote: > BugLink: https://bugs.launchpad.net/bugs/1922678 > > > SRU Justification: > > The patches are fixing a possible memory hog and aging active ct conns. > > * Explain the bug(s) > > Memory bloat and aging active ct conns can happen from any stress test > as we have a single workqueue for handling the ct offload conns > for add/del/stats actions. > > * brief explanation of fixes > > The fix is splitting add/del/stats for diff workqueue and also > we set a limit for add work entries. > > The first patch is already upstream. > > * How to test > > Testing was done with just stress traffic with lots of conns different 5-tuple. > > * What it could break. > > Can take a lot of memory for a long time and also causing active conns to > age out when not necessary. > > > Oz Shlomo (1): > netfilter: flowtable: separate replace, destroy and stats to different > workqueues > > Roi Dayan (1): > UBUNTU: SAUCE: netfilter: nf_flow_table_offload: Limit work entries on > offload add wq > > net/netfilter/nf_flow_table_offload.c | 56 +++++++++++++++++++++++---- > 1 file changed, 48 insertions(+), 8 deletions(-) > Applied to focal:linux-bluefield/master-next. Thanks. -Stefan
BugLink: https://bugs.launchpad.net/bugs/1922678 SRU Justification: The patches are fixing a possible memory hog and aging active ct conns. * Explain the bug(s) Memory bloat and aging active ct conns can happen from any stress test as we have a single workqueue for handling the ct offload conns for add/del/stats actions. * brief explanation of fixes The fix is splitting add/del/stats for diff workqueue and also we set a limit for add work entries. The first patch is already upstream. * How to test Testing was done with just stress traffic with lots of conns different 5-tuple. * What it could break. Can take a lot of memory for a long time and also causing active conns to age out when not necessary. Oz Shlomo (1): netfilter: flowtable: separate replace, destroy and stats to different workqueues Roi Dayan (1): UBUNTU: SAUCE: netfilter: nf_flow_table_offload: Limit work entries on offload add wq net/netfilter/nf_flow_table_offload.c | 56 +++++++++++++++++++++++---- 1 file changed, 48 insertions(+), 8 deletions(-)