[ML][Data Frame] Adding bwc tests for pivot transform #43506
Conversation
benwtrent
added
>test
|
Pinging @elastic/ml-core |
There was a problem hiding this comment.
Can the test have a DF created in the old cluster and left started during upgrade? That way we check DFs left running survive a rolling upgrade. The assertion would be that task state == STARTED but it would be good to check the DF has an assignment. Can the test index more data in the upgraded cluster and restart the DF then assert the stat counts have updated. If that isn't possible in the yml test framework it can be done in Java see MlMappingsUpgradeIT
|
@davidkyle I am not sure there is a reliable way to keep a DF in the |
|
@davidkyle I am attempting to add some BWC tests that have a continuous data frame running while the upgrade is going...hopefully I get it figured out :) |
| SeqNoStats seqNoStats = shard.getSeqNoStats(); | ||
| // SeqNoStats could be `null`. This indicates that an `AlreadyClosed` exception was thrown somewhere down the stack | ||
| // Indicates that the index COULD be closed, or at least that the shard is not fully recovered yet. | ||
| if (seqNoStats == null) { |
There was a problem hiding this comment.
@hendrikmuhs let me know what you think.
It is possible for this to throw an NPE, so to protect against that, I am still throwing a checkpoint exception with the index that failed, and that we were unable to gather seq_no_stats. The user making the call should not see any privileged information.
I opted to log the ShardStats as a warning. If that is too much information, we will have to change this to figure out how to log actionable information given the seq_no_stats being null.
|
@elasticmachine update branch |
* [ML][Data Frame] Adding bwc tests for pivot transform * adding continuous transforms * adding continuous dataframes to bwc * adding continuous data frame tests * Adding rolling upgrade tests for continuous df * Fixing test * Adjusting indices used in BWC, and handling NPE for seq_no_stats * updating and muting specific bwc test
…43929) * [ML][Data Frame] Adding bwc tests for pivot transform (#43506) * [ML][Data Frame] Adding bwc tests for pivot transform * adding continuous transforms * adding continuous dataframes to bwc * adding continuous data frame tests * Adding rolling upgrade tests for continuous df * Fixing test * Adjusting indices used in BWC, and handling NPE for seq_no_stats * updating and muting specific bwc test * Adjusting bwc tests for backport
…43506) (elastic#43929) * [ML][Data Frame] Adding bwc tests for pivot transform (elastic#43506) * [ML][Data Frame] Adding bwc tests for pivot transform * adding continuous transforms * adding continuous dataframes to bwc * adding continuous data frame tests * Adding rolling upgrade tests for continuous df * Fixing test * Adjusting indices used in BWC, and handling NPE for seq_no_stats * updating and muting specific bwc test * Adjusting bwc tests for backport
…43929) (#43947) * [ML][Data Frame] Adding bwc tests for pivot transform (#43506) * [ML][Data Frame] Adding bwc tests for pivot transform * adding continuous transforms * adding continuous dataframes to bwc * adding continuous data frame tests * Adding rolling upgrade tests for continuous df * Fixing test * Adjusting indices used in BWC, and handling NPE for seq_no_stats * updating and muting specific bwc test * Adjusting bwc tests for backport
This adds some new BWC tests for data frame pivot transforms.
When backporting, we will have to black list these tests on clusters where the old version is before 7.2.0.
Additionally, since
pipelineis not supported in 7.2.0, that field will have to be removed from thecomplexold cluster job.Since it is not possible to capture all combinations of parameters efficiently, I opted to have two transforms for the Old, mixed, and new cluster tests. One with minimal viable config, and another with a config that fills out all optional fields.