SHiP: tests for restarting from corrupted log and index files #983
Conversation
tests/ship_restart_test.py
Outdated
| return True | ||
|
|
||
| # Verifies that SHiP should fail to restart with a corrupted first entry header | ||
| def corruptedHeaderTest(pos, curruptedValue, shipNode): |
tests/ship_restart_test.py
Outdated
| stateHistoryIndex = "" | ||
|
|
||
| # Returns True if file1 and file2 contain the same content | ||
| def equalFiles(file1, file2): |
There was a problem hiding this comment.
Can you use Utils.compareFiles instead?
There was a problem hiding this comment.
Thanks. I did not notice that function. I just need to add an extra argument for file mode (the original version does not work for binary file comparison).
…compareFiles() instead of own equalFiles()
|
Note:start |
| Print("Stand up cluster") | ||
|
|
||
| specificExtraNodeosArgs={} | ||
| specificExtraNodeosArgs[shipNodeId]="--plugin eosio::state_history_plugin --trace-history --chain-state-history --finality-data-history --state-history-stride 200 --plugin eosio::net_api_plugin --plugin eosio::producer_api_plugin" |
There was a problem hiding this comment.
Does the chain ever make it past block 200 to exercise anything other then the simple "flat log" case? But I suppose there is probably not much that can be done for a corrupted non-head split log anyways -- fixing it just makes a hole that isn't allowed.
I do wonder if we should do some additional corruption tests with pruned logs. But those are even less repairable.
There was a problem hiding this comment.
Does the chain ever make it past block 200 to exercise anything other then the simple "flat log" case? But I suppose there is probably not much that can be done for a corrupted non-head split log anyways -- fixing it just makes a hole that isn't allowed.
I do wonder if we should do some additional corruption tests with pruned logs. But those are even less repairable.
Thanks. I created #1011 to track additional tests for corrupted pruned and split tests. I think they should be done as a part of the bigger initiative to revisit how corrupted logs are handled (repairing at the startup or offline via spring-util.
I am merging the PR in current form so we have some basic tests.
Added a suite of tests to verify SHiP recovers when restarted in a variety of log and index corruption scenarios.
Resolves #947