jepsen: updated results, confirming that task3 works
All checks were successful
continuous-integration/drone/push Build is passing
continuous-integration/drone/pr Build is passing

This commit is contained in:
Alex 2023-12-14 16:23:48 +01:00
parent 92dd2bbe15
commit fa9247f11b
12 changed files with 96 additions and 31 deletions

View file

@ -13,4 +13,5 @@ pom.xml.asc
.hg/
.direnv
/store
/store.*
.vagrant

View file

@ -7,29 +7,19 @@ Jepsen checking of Garage consistency properties.
Requirements:
- vagrant
- VirtualBox, configured so that nodes can take an IP in a private network `192.168.56.0/24`
- VirtualBox, configured so that nodes can take an IP in a private network `192.168.56.0/24` (it's the default)
- a user that can create VirtualBox VMs
- leiningen
- gnuplot
Set up VMs:
Set up VMs before running tests:
```
vagrant up
```
Run tests (this one should fail):
Run tests: see commands below.
```
lein run test --nodes-file nodes.vagrant --time-limit 64 --concurrency 50 --rate 50 --workload reg
```
These ones are working:
```
lein run test --nodes-file nodes.vagrant --time-limit 64 --rate 50 --concurrency 50 --workload set1
lein run test --nodes-file nodes.vagrant --time-limit 64 --rate 50 --concurrency 50 --workload set2
```
## Results
@ -73,16 +63,19 @@ Results with timestamp patch (`--patch tsfix2`):
Example of a failed run: `garage reg2/20231024T120806.899+0200`.
This is the failure mode we are looking for and trying to fix for NLnet task 3.
- Changes brought by NLnet task 3 code (commit 707442f5de):
no failures with `--scenario r` (0 of 10 runs), `--scenario pr` (0 of 10 runs),
Results with NLnet task 3 code (commit 707442f5de, `--patch task3a`):
- No failures with `--scenario r` (0 of 10 runs), `--scenario pr` (0 of 10 runs),
`--scenario cpr` (0 of 10 runs) and `--scenario dpr` (0 of 10 runs).
- Same with `--patch task3c` (commit `0041b013`, the final version).
### Set, basic test (write some items, then read)
Command: `lein run test --nodes-file nodes.vagrant --time-limit 60 --rate 200 --concurrency 200 --workload set1 --ops-per-key 100 --patch tsfix2`
Command: `lein run test --nodes-file nodes.vagrant --time-limit 60 --rate 200 --concurrency 200 --workload set1 --ops-per-key 100`
Results:
Results without NLnet task3 code (`--patch tsfix2`):
- For now, no failures with clock-scramble nemesis + partition nemesis -> TODO long test run
@ -90,15 +83,22 @@ Results:
- **Fails with the partition + layout reconfiguration nemesis** (`--scenario pr`).
Example of a failed run: `garage set1/20231024T172214.488+0200` (1 failure in 4 runs).
TODO: investigate.
This is the failure mode we are looking for and trying to fix for NLnet task 3.
Results with NLnet task 3 code (commit 707442f5de, `--patch task3a`):
- The tests are buggy and often result in an "unknown" validity status, which
is caused by some requests not returning results during network partitions or
other nemesis-induced broken cluster states. However, when the tests were
able to finish, there were no failures with scenarios `r`, `pr`, `cpr`,
`dpr`.
### Set, continuous test (interspersed reads and writes)
Command: `lein run test --nodes-file nodes.vagrant --time-limit 60 --rate 100 --concurrency 100 --workload set2 --ops-per-key 100 --patch tsfix2`
Command: `lein run test --nodes-file nodes.vagrant --time-limit 60 --rate 100 --concurrency 100 --workload set2 --ops-per-key 100`
Results:
Results without NLnet task3 code (`--patch tsfix2`):
- No failures with clock-scramble nemesis + db nemesis + partition nemesis (`--scenario cdp`) (0 failures in 10 runs).
@ -106,17 +106,26 @@ Results:
Example of a failed run: `garage set2/20231025T141940.198+0200` (10 failures in 10 runs).
This is the failure mode we are looking for and trying to fix for NLnet task 3.
- Changes brought by NLnet task 3 code (commit 707442f5de):
no failures with `--scenario r` (0 of 10 runs), `--scenario pr` (0 of 10 runs).
Results with NLnet task3 code (commit 707442f5de, `--patch task3a`):
- No failures with `--scenario r` (0 of 10 runs), `--scenario pr` (0 of 10 runs),
`--scenario cpr` (0 of 10 runs) and `--scenario dpr` (0 of 10 runs).
- Same with `--patch task3c` (commit `0041b013`, the final version).
## NLnet task 3 final results
- With code from task3 (`--patch task3c`): [reg2 and set2](results/Results-2023-12-13-task3c.png), [set1](results/Results-2023-12-14-task3-set1.png).
- Without (`--patch tsfix2`): [reg2 and set2](results/Results-2023-12-13-tsfix2.png), set1 TBD.
## Investigating (and fixing) errors
### Segfaults
They are due to the download being interrupted in the middle (^C during first launch on clean VMs), the `garage` binary is truncated.
Add `:force?` to the `cached-wget!` call in `daemon.clj` to re-download the binary.
Add `:force?` to the `cached-wget!` call in `daemon.clj` to re-download the binary,
or restar the VMs to clear temporary files.
### In `jepsen.garage`: prefix wierdness

View file

@ -29,4 +29,12 @@ Vagrant.configure("2") do |config|
config.vm.define "n5" do |config| vm(config, "n5", "192.168.56.25") end
config.vm.define "n6" do |config| vm(config, "n6", "192.168.56.26") end
config.vm.define "n7" do |config| vm(config, "n7", "192.168.56.27") end
config.vm.define "n8" do |config| vm(config, "n8", "192.168.56.28") end
config.vm.define "n9" do |config| vm(config, "n9", "192.168.56.29") end
config.vm.define "n10" do |config| vm(config, "n10", "192.168.56.30") end
config.vm.define "n11" do |config| vm(config, "n11", "192.168.56.31") end
config.vm.define "n12" do |config| vm(config, "n12", "192.168.56.32") end
config.vm.define "n13" do |config| vm(config, "n13", "192.168.56.33") end
config.vm.define "n14" do |config| vm(config, "n14", "192.168.56.34") end
end

View file

@ -0,0 +1,18 @@
#!/usr/bin/env bash
set -x
#for ppatch in task3c task3a tsfix2; do
for ppatch in tsfix2; do
#for psc in c cp cdp r pr cpr dpr; do
for psc in cdp r pr cpr dpr; do
#for ptsk in reg2 set1 set2; do
for ptsk in set1; do
for irun in $(seq 10); do
lein run test --nodes-file nodes.vagrant \
--time-limit 60 --rate 100 --concurrency 100 --ops-per-key 100 \
--workload $ptsk --patch $ppatch --scenario $psc
done
done
done
done

View file

@ -0,0 +1,16 @@
#!/usr/bin/env bash
set -x
#for ppatch in task3c tsfix2; do
for ppatch in tsfix2; do
for psc in cdp r pr cpr dpr; do
for ptsk in set1; do
for irun in $(seq 10); do
lein run test --nodes-file nodes2.vagrant \
--time-limit 60 --rate 100 --concurrency 100 --ops-per-key 100 \
--workload $ptsk --patch $ppatch --scenario $psc
done
done
done
done

View file

@ -0,0 +1,7 @@
192.168.56.28
192.168.56.29
192.168.56.30
192.168.56.31
192.168.56.32
192.168.56.33
192.168.56.34

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.4 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1.1 MiB

Binary file not shown.

After

Width:  |  Height:  |  Size: 1 MiB

View file

@ -36,7 +36,9 @@
{"default" "v0.9.0"
"tsfix1" "d146cdd5b66ca1d3ed65ce93ca42c6db22defc09"
"tsfix2" "c82d91c6bccf307186332b6c5c6fc0b128b1b2b1"
"task3a" "707442f5de416fdbed4681a33b739f0a787b7834"})
"task3a" "707442f5de416fdbed4681a33b739f0a787b7834"
"task3b" "431b28e0cfdc9cac6c649193cf602108a8b02997"
"task3c" "0041b013a473e3ae72f50209d8f79db75a72848b"})
(def cli-opts
"Additional command line options."
@ -69,7 +71,7 @@
(merge tests/noop-test
opts
{:pure-generators true
:name (str "garage " (name (:workload opts)) " " (name (:scenario opts)) " " (name (:patch opts)))
:name (str "garage-" (name (:patch opts)) " " (name (:workload opts)) " " (name (:scenario opts)))
:os debian/os
:db db
:client (:client workload)
@ -83,7 +85,9 @@
(gen/nemesis (:final-generator scenario))
(gen/log "Waiting for recovery")
(gen/sleep 10)
(gen/clients (:final-generator workload)))
(gen/log "Running final generator")
(gen/clients (:final-generator workload))
(gen/log "Generators all done"))
:nemesis (:nemesis scenario)
:checker (checker/compose
{:perf (checker/perf (:perf scenario))

View file

@ -108,11 +108,13 @@
(->> (range)
(map (fn [x] {:type :invoke, :f :add, :value x}))
(gen/limit (:ops-per-key opts)))))
:final-generator (gen/phases
(independent/sequential-generator
:final-generator (independent/concurrent-generator
10
(range 100)
(fn [k] (gen/once op-read)))
(gen/sleep 5))})
(fn [k]
(gen/phases
(gen/once op-read)
(gen/sleep 5))))})
(defn workload2
"Tests insertions and deletions"