Xiaoxi Chen

Results 11 issues of Xiaoxi Chen

When the allocator size is small, it is possible we have a size 0 level in slab, which will crash in constructing SlabCacheQueue

https://jubianchi.github.io/semver-check/#/^12.1/12.4

``` for (const auto& slab_cfg : cfg.m_per_slab_cfg) { std::vector< blk_num_t > level_limits; level_limits.reserve(slab_cfg.m_level_distribution_pct.size()); #ifndef NDEBUG HS_DBG_ASSERT_EQ(slab_cfg.slab_size, slab_size, "Slab config size is not contiguous power of 2"); slab_size *= 2; #endif...

``` [03/10/24 21:06:21-07:00] [I] [508688] [replication_state_machine.cpp:7:on_commit] applying raft log commit with lsn:232 [03/10/24 21:06:21-07:00] [I] [508141] [hs_blob_tests.cpp:73:TestBody] delete blob shard 1125899906842625 blob 51 [03/10/24 21:06:21-07:00] [I] [508141] [replication_state_machine.cpp:35:on_pre_commit] on_pre_commit with...

bug

log ``` [03/01/24 00:30:23-07:00] [D] [397827] [index_kv.cpp:70:get_blob_from_index_table] Failed to get from index table [route=0001:000000000001:0000000000000000] [03/01/24 00:30:23-07:00] [W] [397827] [hs_blob_manager.cpp:207:_get_blob] [pg=1,shard=1,blob=0] Blob not found in index [03/01/24 00:30:23-07:00] [D] [397827] [fixture_app.cpp:57:SetUp]...

bug

With this we can easily get time diff of each given state, avoid doing the calculation/analysis from grep of logs.

he m_resync_mode is always to false which lead each resync go through general IO path --- we queue the fetch data requests to a queue, the reaper_thread picks from the...

bug

majar changes: 1. fix device list 2. when leader switches , new leader will continue to write in `write_on_leader`. 3. logging improvement, print out ms/us for timestamp. Other than UT...

This is to ensure data written to the vdev(and fiannly to the pdev), is actually written to disk, even pdev uses buffer io.

[stack.txt](https://github.com/user-attachments/files/17001693/stack.txt) In sm long running we tried to kill the process with sigterm, the process is hanging afterward and cannot exit sometimes. call stack attached. Seems like someone is holing...

bug