store: use k-way merging for proxying logic #5296

GiedriusS · 2022-04-24T09:47:08Z

Use k-way merging for merging multiple responses coming from multiple
StoreAPIs. This avoids having a bottleneck of a Go channel that can only
hold 10 responses at once. So, now we will Recv() messages as fast as
possible. Users should be able to see much quicker queries because
network will be used as much as possible.

Benchmarks:

name                                                                                     old time/op    new time/op    delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        1.01s ± 7%     0.24s ± 2%  -76.62%  (p=0.008 n=5+5)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       926ms ± 3%     454ms ± 2%  -50.99%  (p=0.008 n=5+5)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16      95.6ms ± 1%    23.4ms ± 3%  -75.52%  (p=0.016 n=4+5)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16    93.1ms ± 2%    45.1ms ± 1%  -51.63%  (p=0.008 n=5+5)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16     39.5µs ± 1%    25.2µs ± 1%  -36.22%  (p=0.008 n=5+5)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16    22.6µs ± 0%    15.2µs ± 0%  -32.71%  (p=0.008 n=5+5)

name                                                                                     old alloc/op   new alloc/op   delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        186MB ± 0%     108MB ± 0%  -41.87%  (p=0.008 n=5+5)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       186MB ± 0%     107MB ± 0%  -42.18%  (p=0.008 n=5+5)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16      18.5MB ± 0%    10.0MB ± 0%  -45.79%  (p=0.008 n=5+5)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16    18.5MB ± 0%    10.6MB ± 0%  -42.69%  (p=0.008 n=5+5)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16     10.9kB ± 0%     6.1kB ± 0%  -44.06%  (p=0.008 n=5+5)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16    5.09kB ± 0%    3.06kB ± 0%  -39.83%  (p=0.008 n=5+5)

name                                                                                     old allocs/op  new allocs/op  delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        5.00M ± 0%     1.00M ± 0%  -80.00%  (p=0.008 n=5+5)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       5.00M ± 0%     1.00M ± 0%  -80.00%  (p=0.008 n=5+5)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16        500k ± 0%      100k ± 0%  -79.96%  (p=0.008 n=5+5)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16      500k ± 0%      100k ± 0%  -79.98%  (p=0.016 n=5+4)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16        180 ± 0%       125 ± 0%  -30.56%  (p=0.008 n=5+5)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16       109 ± 0%        80 ± 0%  -26.61%  (p=0.008 n=5+5)

TestProxyStore_SeriesSlowStores also passes which tests how everything works with slow stores.

Rewritten chunk deduplication logic to use hashing.

In the future, we could get some more marginal gains by using a selection/tournament tree.

Signed-off-by: Giedrius Statkevičius giedrius.statkevicius@vinted.com

Use k-way merging for merging multiple responses coming from multiple StoreAPIs. This avoids having a bottleneck of a Go channel that can only hold 10 responses at once. So, now we will Recv() messages as fast as possible. Users should be able to see much quicker queries because network will be used as much as possible. Benchmarks: ``` name old time/op new time/op delta ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16 1.01s ± 7% 0.24s ± 2% -76.62% (p=0.008 n=5+5) ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16 926ms ± 3% 454ms ± 2% -50.99% (p=0.008 n=5+5) ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16 95.6ms ± 1% 23.4ms ± 3% -75.52% (p=0.016 n=4+5) ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16 93.1ms ± 2% 45.1ms ± 1% -51.63% (p=0.008 n=5+5) ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16 39.5µs ± 1% 25.2µs ± 1% -36.22% (p=0.008 n=5+5) ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16 22.6µs ± 0% 15.2µs ± 0% -32.71% (p=0.008 n=5+5) name old alloc/op new alloc/op delta ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16 186MB ± 0% 108MB ± 0% -41.87% (p=0.008 n=5+5) ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16 186MB ± 0% 107MB ± 0% -42.18% (p=0.008 n=5+5) ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16 18.5MB ± 0% 10.0MB ± 0% -45.79% (p=0.008 n=5+5) ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16 18.5MB ± 0% 10.6MB ± 0% -42.69% (p=0.008 n=5+5) ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16 10.9kB ± 0% 6.1kB ± 0% -44.06% (p=0.008 n=5+5) ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16 5.09kB ± 0% 3.06kB ± 0% -39.83% (p=0.008 n=5+5) name old allocs/op new allocs/op delta ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16 5.00M ± 0% 1.00M ± 0% -80.00% (p=0.008 n=5+5) ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16 5.00M ± 0% 1.00M ± 0% -80.00% (p=0.008 n=5+5) ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16 500k ± 0% 100k ± 0% -79.96% (p=0.008 n=5+5) ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16 500k ± 0% 100k ± 0% -79.98% (p=0.016 n=5+4) ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16 180 ± 0% 125 ± 0% -30.56% (p=0.008 n=5+5) ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16 109 ± 0% 80 ± 0% -26.61% (p=0.008 n=5+5) ``` `TestProxyStore_SeriesSlowStores` also passes which tests how everything works with slow stores. Rewritten chunk deduplication logic to use hashing. In the future we could get some more marginal gains by using a selection/tournament tree. Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

pkg/store/storepb/custom.go

pkg/store/proxy_heap.go

bwplotka · 2022-04-25T15:38:05Z

This is pretty sweet 💪🏽 Sounds like this might fix #4780 - will take a look this week or next (:

pkg/store/proxy_heap.go

GiedriusS · 2022-05-02T07:43:01Z

Tried this pull request in prod with a query that returns ~21k series - it reduced the query duration by about 30% 🤯

MD5 is fine here because we don't care as much about collisions. Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

pkg/store/proxy_heap.go

fpetkovski · 2022-05-05T12:56:11Z

I tested this in our infra and can confirm improved performance for queries across lots of series. Some queries went down from 40s to 30s. Dashboards also seem to be loading more smoothly.

Looking forward to seeing this merged 🚀

bwplotka

Thanks @GiedriusS for leading this work, amazing stuff!

I think we could switch to this, but there are two points to remember:

I think the efficiency boost (e.g latency) is caused not by a different merge algorithm, but actually by buffering all messages ASAP (which we agreed makes sense)

We could trigger an infinite (kind of) buffer by just changing this channel length to something large (e.g million).. which is obv bad, because we would always allocate that many.

My point is: In this PR with buffering all messages change (which is amazing), we ALSO changed to k-merge which might be actually slower/worse than the previous MergeSort. We don't know as we changed two things in the same time here.

We still have a bottleneck on waiting for ALL messages until we start streaming merged responses. I think we can do k-merge in more asynchronous way. My worry is that if we want to implement it we will have to refactor this code 3rd time 🙃

Maybe those 2 things are fine to agree to. I would be keen on merging this as it's simple and have immediate value, so approving. I added some readability suggestions (nits).

pkg/store/proxy.go

pkg/store/proxy_heap.go

pkg/store/proxy_test.go

bwplotka

Well, I think we also have to figure out this: #5296 (comment)

This feels like a future problem if we ignore it 🤔

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

bwplotka

Nice! Amazing. Thanks for making lazy response buffering 👍🏽

Wonder what's the benchmarks are now. Happy to merge!

GiedriusS · 2022-05-10T21:34:21Z

Please don't merge this yet - even though the code works and the benchmarks look even better (will post tomorrow) but I think there's some lock contention going. So, let me come back with some final tweaks and a bigger comment.

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

GiedriusS · 2022-06-06T09:21:54Z

Mhm, it seems like it is impossible to implement this properly because gRPC RecvMsg() doesn't take any context variable, it uses the stream's context. Thus, to implement a per-request timeout, one needs to spawn a separate goroutine. This implies that one will need a channel in the middle to send responses back to the original goroutine or you'll need to synchronize these two things somehow differently. Because our goal is to RecvMsg() as fast as possible, it means that we must avoid having a channel in the middle. We could do a simple loop recv -> append in the main goroutine but if RecvMsg() blocks indefinitely then it would block everything going after it so two separate receiver/sender goroutines are necessary. grpc/grpc-go#1229 (comment) this is a real pity that this is the accepted solution, to spawn a new goroutine for each RecvMsg() operation. I'll try to think about this more.

I have filed grpc/grpc-go#5401 in the mean-time.

bwplotka · 2022-06-23T08:18:57Z

Any progress?

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

GiedriusS · 2022-09-01T14:27:20Z

Updated PR:

Fixed everything according to comments;
Remove Error() method to not mislead readers - we actually convert all errors into warnings now and then the Series() function itself decides whether those warnings are errors or not depending on what partial response strategy has been set;
Added a hidden parameter for setting the proxying strategy and used it in e2e tests to test how it works;
Tried to repro @fpetkovski's case in e2e tests but cannot reproduce it;
Brought back could not receive data from X StoreAPI in Y seconds - we now check whether the timer has fired or not if the error is context.Canceled. If that's true then it means that the timer has fired and that the per-Recv() timeout has been reached.

Eager strategy benchmarks:

name                                                                                     old time/op    new time/op    delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        1.03s ± 5%     0.18s ± 5%   -82.34%  (p=0.000 n=10+10)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       955ms ± 4%     260ms ±14%   -72.76%  (p=0.000 n=10+10)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16      96.2ms ± 3%    18.7ms ± 5%   -80.54%  (p=0.000 n=8+9)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16     117ms ±22%      25ms ± 2%   -79.03%  (p=0.000 n=10+8)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16     42.4µs ± 9%    30.8µs ± 8%   -27.42%  (p=0.000 n=10+10)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16    23.6µs ± 4%    17.6µs ± 2%   -25.42%  (p=0.000 n=10+10)

name                                                                                     old alloc/op   new alloc/op   delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        186MB ± 0%      84MB ± 0%   -54.79%  (p=0.000 n=9+10)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       186MB ± 0%      83MB ± 0%   -55.11%  (p=0.000 n=9+10)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16      18.5MB ± 0%     7.6MB ± 0%   -58.75%  (p=0.000 n=10+10)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16    18.5MB ± 0%     8.2MB ± 0%   -55.66%  (p=0.000 n=9+10)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16     11.3kB ± 0%     8.1kB ± 0%   -28.10%  (p=0.000 n=9+9)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16    5.20kB ± 0%    3.47kB ± 0%   -33.15%  (p=0.000 n=10+10)

name                                                                                     old allocs/op  new allocs/op  delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        5.00M ± 0%     0.00M ± 1%   -99.99%  (p=0.000 n=9+10)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       5.00M ± 0%     0.00M ± 1%  -100.00%  (p=0.000 n=10+10)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16        500k ± 0%        0k ± 0%   -99.95%  (p=0.000 n=10+8)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16      500k ± 0%        0k ± 0%   -99.97%  (p=0.000 n=9+9)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16        185 ± 0%       144 ± 0%   -22.16%  (p=0.000 n=10+10)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16       111 ± 0%        81 ± 0%   -27.03%  (p=0.000 n=10+10)

Lazy strategy benchmarks:

name                                                                                     old time/op    new time/op    delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        1.03s ± 5%     0.19s ± 3%  -81.92%  (p=0.000 n=10+10)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       955ms ± 4%     274ms ± 2%  -71.29%  (p=0.000 n=10+10)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16      96.2ms ± 3%    19.3ms ± 5%  -79.89%  (p=0.000 n=8+10)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16     117ms ±22%      28ms ± 2%  -76.16%  (p=0.000 n=10+9)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16     42.4µs ± 9%    30.4µs ± 1%  -28.33%  (p=0.000 n=10+9)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16    23.6µs ± 4%    17.3µs ± 3%  -26.50%  (p=0.000 n=10+10)

name                                                                                     old alloc/op   new alloc/op   delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        186MB ± 0%      81MB ± 1%  -56.17%  (p=0.000 n=9+10)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       186MB ± 0%      77MB ± 5%  -58.43%  (p=0.000 n=9+10)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16      18.5MB ± 0%     7.4MB ± 1%  -60.27%  (p=0.000 n=10+10)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16    18.5MB ± 0%     6.9MB ± 4%  -62.51%  (p=0.000 n=9+10)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16     11.3kB ± 0%     8.4kB ± 0%  -25.54%  (p=0.000 n=9+9)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16    5.20kB ± 0%    3.55kB ± 0%  -31.77%  (p=0.000 n=10+10)

name                                                                                     old allocs/op  new allocs/op  delta
ProxySeries/1000000SeriesWith1Samples/4_client_with_1_samples,_250000_series_each-16        5.00M ± 0%     0.00M ±52%  -99.99%  (p=0.000 n=9+9)
ProxySeries/1000000SeriesWith1Samples/single_client_with_1_samples,_1000000_series-16       5.00M ± 0%    0.00M ±107%  -99.98%  (p=0.000 n=10+9)
ProxySeries/100000SeriesWith100Samples/4_client_with_25_samples,_25000_series_each-16        500k ± 0%        0k ±13%  -99.93%  (p=0.000 n=10+10)
ProxySeries/100000SeriesWith100Samples/single_client_with_100_samples,_100000_series-16      500k ± 0%        0k ±72%  -99.91%  (p=0.000 n=9+10)
ProxySeries/1SeriesWith10000000Samples/4_client_with_2500000_samples,_1_series_each-16        185 ± 0%       148 ± 0%  -20.00%  (p=0.000 n=10+10)
ProxySeries/1SeriesWith10000000Samples/single_client_with_10000000_samples,_1_series-16       111 ± 0%        82 ± 0%  -26.13%  (p=0.000 n=10+10)

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

fpetkovski

I am happy to see this merged in the current state. The issue I noticed might be specific to our environment, so if I can troubleshoot it myself if it happens again.

fpetkovski · 2022-09-05T05:04:58Z

Ha, the e2e test failure seems to be exactly what I saw :)

GiedriusS · 2022-09-06T09:11:35Z

Ha, the e2e test failure seems to be exactly what I saw :)

Yeah 👍 trying to fix it and then we can merge this

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

GiedriusS · 2022-09-08T13:48:50Z

I think I have fixed all of the bugs. PTAL @bwplotka @fpetkovski @yeya24 🙏

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

fpetkovski · 2022-09-09T10:36:53Z

pkg/store/proxy_heap.go

+				if field == nil {
+					continue
+				}
+				h := xxhash.Sum64(field.Data)


As a follow up improvement, would it make sense to do this hashing in the stores so that we offload some work from the querier proxy? We might also think about other operations that can be pre-computed at the store level.

bwplotka

I love it! @fpetkovski also had amazing improvements already with eager option 💪🏽

Thanks for addressing all. Let's go! 👍🏽

It's normal and not an error if a query does not match due to no downstream stores. This is common when querying with external labels and tiered query servers. This bug was introduced in thanos-io#5296 Fixes: thanos-io#5862 Signed-off-by: SuperQ <superq@gmail.com>

It's normal and not an error if a query does not match due to no downstream stores. This is common when querying with external labels and tiered query servers. This bug was introduced in #5296 Fixes: #5862 Signed-off-by: SuperQ <superq@gmail.com>

It's normal and not an error if a query does not match due to no downstream stores. This is common when querying with external labels and tiered query servers. This bug was introduced in thanos-io#5296 Fixes: thanos-io#5862 Signed-off-by: SuperQ <superq@gmail.com>

pull-request-size bot added the size/XL label Apr 24, 2022

GiedriusS force-pushed the kway_merge_heap branch from d697c8d to 360342f Compare April 24, 2022 10:02

GiedriusS force-pushed the kway_merge_heap branch from 360342f to 0fb4a18 Compare April 24, 2022 11:09

Merge remote-tracking branch 'origin/main' into kway_merge_heap

8fc05fe

GiedriusS marked this pull request as ready for review April 24, 2022 12:23

GiedriusS requested a review from bwplotka April 24, 2022 12:23

yeya24 reviewed Apr 24, 2022

View reviewed changes

pkg/store/storepb/custom.go Outdated Show resolved Hide resolved

pkg/store/proxy_heap.go Outdated Show resolved Hide resolved

fpetkovski reviewed Apr 25, 2022

View reviewed changes

pkg/store/proxy_heap.go Outdated Show resolved Hide resolved

store: reuse hash + use md5

0eda550

MD5 is fine here because we don't care as much about collisions. Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

sonatype-lift bot reviewed May 3, 2022

View reviewed changes

pkg/store/proxy_heap.go Outdated Show resolved Hide resolved

sonatype-lift bot reviewed May 3, 2022

View reviewed changes

pkg/store/proxy_heap.go Outdated Show resolved Hide resolved

bwplotka mentioned this pull request May 5, 2022

Querier: Deterministic buffered queries #4780

Closed

bwplotka approved these changes May 5, 2022

View reviewed changes

bwplotka requested changes May 5, 2022

View reviewed changes

GiedriusS added 2 commits May 10, 2022 13:39

store: add full streaming to proxying logic

666ed17

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

Merge remote-tracking branch 'origin/main' into kway_merge_heap

244f2fc

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

bwplotka previously approved these changes May 10, 2022

View reviewed changes

GiedriusS dismissed bwplotka’s stale review via fc4cd7a May 30, 2022 15:00

pull-request-size bot added size/XXL and removed size/XL labels May 30, 2022

GiedriusS added 2 commits May 30, 2022 18:01

proxy: make strategy tunable

fc4cd7a

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

proxy: fix off-by-1

c3532dd

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

GiedriusS marked this pull request as draft June 6, 2022 09:22

GiedriusS added 2 commits September 1, 2022 17:00

query: remove unused constants

07f680f

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

store: wire up proxy strategy

9ce4e0f

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

e2e: add repro for Filip's case

c0d3f46

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

fpetkovski previously approved these changes Sep 5, 2022

View reviewed changes

store: fix lazyRespSet At()

003c51e

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

GiedriusS dismissed fpetkovski’s stale review via 003c51e September 8, 2022 12:36

Merge remote-tracking branch 'origin/main' into kway_merge_heap

2d28e05

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

storepb: remove unused method

caad5c3

Signed-off-by: Giedrius Statkevičius <giedrius.statkevicius@vinted.com>

fpetkovski reviewed Sep 9, 2022

View reviewed changes

bwplotka approved these changes Sep 9, 2022

View reviewed changes

bwplotka merged commit 32a6c4e into thanos-io:main Sep 9, 2022

GiedriusS deleted the kway_merge_heap branch September 9, 2022 10:49

GiedriusS mentioned this pull request Sep 9, 2022

proxy: pre-calculate hash at leaf nodes #5682

Closed

fpetkovski mentioned this pull request Sep 24, 2022

Improve detections for sorting data before dedup #5719

Open

GiedriusS mentioned this pull request Nov 4, 2022

Upgrade to v2.28.1 - > v2.29.0 An error was reported #5862

Closed

SuperQ mentioned this pull request Jan 29, 2023

Store: Don't error when no stores are matched #6082

Merged

2 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

store: use k-way merging for proxying logic #5296

store: use k-way merging for proxying logic #5296

GiedriusS commented Apr 24, 2022

bwplotka commented Apr 25, 2022

GiedriusS commented May 2, 2022

fpetkovski commented May 5, 2022 •

edited

Loading

bwplotka left a comment •

edited

Loading

bwplotka left a comment

bwplotka left a comment

GiedriusS commented May 10, 2022

GiedriusS commented Jun 6, 2022 •

edited

Loading

bwplotka commented Jun 23, 2022

GiedriusS commented Sep 1, 2022

fpetkovski left a comment

fpetkovski commented Sep 5, 2022

GiedriusS commented Sep 6, 2022

GiedriusS commented Sep 8, 2022

fpetkovski Sep 9, 2022

bwplotka left a comment

store: use k-way merging for proxying logic #5296

store: use k-way merging for proxying logic #5296

Conversation

GiedriusS commented Apr 24, 2022

bwplotka commented Apr 25, 2022

GiedriusS commented May 2, 2022

fpetkovski commented May 5, 2022 • edited Loading

bwplotka left a comment • edited Loading

Choose a reason for hiding this comment

bwplotka left a comment

Choose a reason for hiding this comment

bwplotka left a comment

Choose a reason for hiding this comment

GiedriusS commented May 10, 2022

GiedriusS commented Jun 6, 2022 • edited Loading

bwplotka commented Jun 23, 2022

GiedriusS commented Sep 1, 2022

fpetkovski left a comment

Choose a reason for hiding this comment

fpetkovski commented Sep 5, 2022

GiedriusS commented Sep 6, 2022

GiedriusS commented Sep 8, 2022

fpetkovski Sep 9, 2022

Choose a reason for hiding this comment

bwplotka left a comment

Choose a reason for hiding this comment

fpetkovski commented May 5, 2022 •

edited

Loading

bwplotka left a comment •

edited

Loading

GiedriusS commented Jun 6, 2022 •

edited

Loading