I had a look at patches 1-8. I have no additional notes beyond what Han
and Ales have already mentioned.
For patches 1-8:
Acked-by: Mark Michelson <[email protected]>
On 8/18/23 04:56, [email protected] wrote:
From: Numan Siddique <[email protected]>
This patch series adds the support to handle load balancer and
load balancer group changes incrementally in the "northd" engine
node and "lflow" engine node. Changes to logical switches and router's load
balancer and load balancer group columns are also handled incrementally
provided other columns do not change.
V4 of this series did not include LB I-P handling in the lflow engine
node. V5 adds 6 more patches to handle the LB changes in the lflow
engine node. V6 added 2 additional patches to handle router NAT I-P
handling.
This patch series can be divided into 3 parts
Part 1. Patches 1 to 8 (LB I-P only in northd)
Part 2. Patches 9 to 14 (LB I-P in lflow engine too)
Part 3. Patches 15 and 16 (LR NAT I-P handling in both northd and
lflow)
Submitting all these patches as one series. These patches can also be
found here - https://github.com/numansiddique/ovn/tree/northd_ip_lb_ip_v6
If there are any conflicts, request the reviewers to use this branch
for reviewing.
Below are the scale testing results done with these patches applied
using ovn-heater. The test ran the scenario -
ocp-500-density-heavy.yml [1].
With these patches applied (with load balancer I-P handling in both
northd and lflow engine nodes) the resuts (Result 1) are:
-------------------------------------------------------------------------------------------------------------------------------------------------------
Min (s) Median (s) 90%ile (s) 99%ile
(s) Max (s) Mean (s) Total (s) Count Failed
-------------------------------------------------------------------------------------------------------------------------------------------------------
Iteration Total 0.135651 1.130527 1.179357
1.201410 2.180203 0.674606 84.325717 125 0
Namespace.add_ports 0.005218 0.005678 0.006457
0.018936 0.020812 0.006182 0.772796 125 0
WorkerNode.bind_port 0.033631 0.043287 0.051171
0.058223 0.062819 0.043839 10.959757 250 0
WorkerNode.ping_port 0.005460 0.006791 1.041434
1.064807 1.069957 0.274352 68.587878 250 0
-------------------------------------------------------------------------------------------------------------------------------------------------------
With only the first 8 patches applied (with load balancer I-P handling
only in northd engine node) the results (Result 2) are:
-------------------------------------------------------------------------------------------------------------------------------------------------------
Min (s) Median (s) 90%ile (s) 99%ile
(s) Max (s) Mean (s) Total (s) Count Failed
-------------------------------------------------------------------------------------------------------------------------------------------------------
Iteration Total 0.132929 2.157103 3.314847
3.331561 4.378626 1.581889 197.736147 125 0
Namespace.add_ports 0.005217 0.005760 0.006565
0.013348 0.021014 0.006106 0.763214 125 0
WorkerNode.bind_port 0.035205 0.045458 0.052278
0.059804 0.063941 0.045652 11.413122 250 0
WorkerNode.ping_port 0.005075 0.006814 3.088548
3.192577 4.242026 0.726453 181.613284 250 0
-------------------------------------------------------------------------------------------------------------------------------------------------------
The results with the present main (Result 3) are:
-------------------------------------------------------------------------------------------------------------------------------------------------------
Min (s) Median (s) 90%ile (s) 99%ile
(s) Max (s) Mean (s) Total (s) Count Failed
-------------------------------------------------------------------------------------------------------------------------------------------------------
Iteration Total 4.377260 6.486962 7.502040
8.322587 8.334701 6.559002 819.875306 125 0
Namespace.add_ports 0.005112 0.005484 0.005953
0.009153 0.011452 0.005662 0.707752 125 0
WorkerNode.bind_port 0.035360 0.042732 0.049152
0.053698 0.056635 0.043215 10.803700 250 0
WorkerNode.ping_port 0.005338 1.599904 7.229649
7.798039 8.206537 3.209860 802.464911 250 0
-------------------------------------------------------------------------------------------------------------------------------------------------------
Few observations:
- The total time taken has come down significantly from 819 seconds to 197
to complete the density heavy tests (excluding the base cluster
bringup) with only the northd engine (Result 2) handling
and it came further down to around 84 seconds with all the patches
applied (Result 1)
- 99%ile with these patches is 3.3 seconds in Result 2 and 1.06 seconds in
Result 1
compared to 8.3 seconds for the main (Result 3).
- 90%file with these patches is 3.3 seconds in Result 2 and 1.04
seconds in Result 1 compared to 7.5 seconds for the main (Result 3).
- CPU utilization of northd during the test with these patches
is between 100% to 300% which is almost the same as main.
Main difference being that, with these patches the test duration is
less and hence overall less CPU utilization.
[1] -
https://github.com/ovn-org/ovn-heater/blob/main/test-scenarios/ocp-500-density-heavy.yml
v5 -> v6
-------
* Rebased. Added 2 more patches (p15 and p16) for LR NAT I-P handling.
v4 -> v5
-------
* 6 new patches are added to the series which handles the LB changes
in the lflow engine node.
v3 -> v4
-------
* Covered more test scearios.
* Found few issues and fixed them. v3 was not handling the scenario of
a vip getting added or removed from a load balancer.
v2 -> v3
--------
* v2 was very inefficient in handling the load balancer group changes
and in associating the load balancers of the lb group to the
datapaths. This was the main reason for the regression in the full
recompute time taken.
v3 addressed these by more efficiently handling the lb group changes
incrementally.
Numan Siddique (16):
northd I-P: Sync SB load balancers in a separate engine node.
northd: Add a new engine node - lb_data.
northd: Add initial I-P for load balancer and load balancer groups
northd: Refactor the 'northd' node code which handles logical switch
changes.
northd: Handle load balancer changes for a logical switch.
northd: Handle load balancer group changes for a logical switch.
northd: Sync SB Port bindings NAT column in a separate engine node.
northd: Handle load balancer/group changes for a logical router.
northd: Use objdep mgr for lport to lflow references.
northd: Fix LSP incremental processing if dhcp options are set.
northd: Use objdep mgr for datapath/lb to lflow references.
Reference lb related lflows for lports in a separate objdep mgr type.
northd: Refactor the northd change tracking.
northd: Handle load balancer change in lflow engine.
northd: Move router ports SB PB options sync to sync_to_sb_pb node.
northd: Handle NAT changes for a logical router incrementally.
lib/lb.c | 320 +-
lib/lb.h | 105 +-
lib/objdep.h | 6 +
lib/ovn-util.c | 11 +-
northd/automake.mk | 2 +
northd/en-lb-data.c | 800 +++++
northd/en-lb-data.h | 109 +
northd/en-lflow.c | 33 +-
northd/en-northd.c | 144 +-
northd/en-northd.h | 5 +
northd/en-sync-sb.c | 76 +
northd/en-sync-sb.h | 10 +
northd/inc-proc-northd.c | 39 +-
northd/northd.c | 6969 +++++++++++++++++++++++++-------------
northd/northd.h | 149 +-
northd/ovn-northd.c | 4 +
tests/ovn-northd.at | 869 +++++
17 files changed, 7145 insertions(+), 2506 deletions(-)
create mode 100644 northd/en-lb-data.c
create mode 100644 northd/en-lb-data.h
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
