Skip to content

compact: relax index label checking#848

Merged
bwplotka merged 1 commit intothanos-io:masterfrom
GiedriusS:relax_compactor
Mar 5, 2019
Merged

compact: relax index label checking#848
bwplotka merged 1 commit intothanos-io:masterfrom
GiedriusS:relax_compactor

Conversation

@GiedriusS
Copy link
Member

@GiedriusS GiedriusS commented Feb 15, 2019

It is possible that Prometheus will produce time series with duplicate
labels. E.g.: a{b="2",b="2"} 1

Where such metric is pushed to Prometheus either wittingly or
unwittingly, the compact process will stall because the check will not
pass. For example:

{"caller":"main.go:181","err":"error executing compaction: compaction failed: compaction: gather index issues for block /data/compact/0@{monitor="monitor",replica="repl"}/01D34EDQMSQ29RHAC47XGKHGC7: out-of-order label set {name="foo",exported_job="vv",host="172_16_226_56",host="172_16_226_56",region="lt",subtask_index="5",task_attempt_id="32e4b047bb768583ff57c709be3b1046",task_attempt_num="8",task_id="688c028a219ff3372f3eecb0ee5811f9",task_name="Source:_foo",tenant="abc",tier="cooltier",tm_id="53b2ed987b08f427dec4ee1465df91fa"} for series 2594231","level":"error","msg":"running command failed","ts":"2019-02-11T13:30:33.901722306Z"}

This commit fixes the mentioned issue. Amend the e2e test to test the
case when a series has identical labels.

Brings us closer to accepting data that Prometheus can produce. Maybe we are still missing something as evident by other issues?! :)

Changes

Relaxed the labels check in the GatherIndexIssueStats function so that it would be permitted to have identical labels.

Verification

Tests pass.

It is possible that Prometheus will produce time series with duplicate
labels. E.g.: a{b="2",b="2"} 1

Where such metric is pushed to Prometheus either wittingly or
unwittingly, the compact process will stall because the check will not
pass. For example:

{"caller":"main.go:181","err":"error executing compaction: compaction failed: compaction: gather index issues for block /data/compact/0@{monitor="monitor",replica="repl"}/01D34EDQMSQ29RHAC47XGKHGC7: out-of-order label set {**name**="foo",exported_job="vv",host="172_16_226_56",host="172_16_226_56",region="lt",subtask_index="5",task_attempt_id="32e4b047bb768583ff57c709be3b1046",task_attempt_num="8",task_id="688c028a219ff3372f3eecb0ee5811f9",task_name="Source:_foo",tenant="abc",tier="cooltier",tm_id="53b2ed987b08f427dec4ee1465df91fa"} for series 2594231","level":"error","msg":"running command failed","ts":"2019-02-11T13:30:33.901722306Z"}

This commit fixes the mentioned issue. Amend the e2e test to test the
case when a series has identical labels.
Copy link
Member

@bwplotka bwplotka left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, let me know if that makes sense, otherwise we will merge.

l0 := lset[0]
for _, l := range lset[1:] {
if l.Name <= l0.Name {
if l.Name < l0.Name {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we at least warn?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You mean warn the user if they are equal? What message do you think would fit here?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yea, for example "Invalid labelset found; multiple labels with the same name for the same series found"

Wonder how is that even possible? (: I would go even further and put the whole verification step under some flag accept-malformed-index, but not sure. (: If any, we can do it in later PRs.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Will merge this, let's think about this later on.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants