~measurement-factory/squid/collapsed-fwd

From Revision 12591 to 12572

expand all

collapse all

Rev		Summary	Authors	Date	Diff	Files
12591		Made "make check" happier by syncing test case with Store AP... Made "make check" happier by syncing test case with Store API changes and making StoreEntry::hashDelete() more forgiving.	Alex Rousskov	10 years ago
12590		Add missing StoreEntry::lock() parameters, require unlock() ... Add missing StoreEntry::lock() parameters, require unlock() parameter, and polish un/lock() context debugging.	Alex Rousskov	10 years ago
12589		Added StoreEntry::lock() call dropped when merging from trun... Added StoreEntry::lock() call dropped when merging from trunk r13172.	Alex Rousskov	10 years ago
12588		Merged from trunk 13172. Merged from trunk 13172.	Alex Rousskov	10 years ago
12587		Fixed src/tests/testRock build. Fixed src/tests/testRock build.	Alex Rousskov	10 years ago
12586		Fixed swap_file_sz calculation when loading rock entries. Po... Fixed swap_file_sz calculation when loading rock entries. Polished debugging. Supply storeRebuildParseEntry() with known "swap file size" so that it can adjust swap_file_sz after loading store entry meta info.  Entries are often stored with swap_file_sz in the meta header missing the swap_hdr_len component. storeRebuildParseEntry() adds swap_hdr_len when needed, using known entry size to detect that need.	Alex Rousskov	10 years ago
12585		Do not use StoreEntry::swap_file_sz to write DbCellHeader::e... Do not use StoreEntry::swap_file_sz to write DbCellHeader::entrySize, even during the last write. StoreEntry::swap_file_sz is often set by storeSwapOutFileClosed, which is called after the last write. Also, I am not sure whether partial StoreEntry::swap_file_sz info might later confuse store rebuild code into thinking that the whole entry is malformed. That would be [different] bug.	Alex Rousskov	10 years ago
12584		Fixed DbCellHeader sanity check: Payload may occupy all spac... Fixed DbCellHeader sanity check: Payload may occupy all space after the header.	Alex Rousskov	10 years ago
12583		Merged from trunk r12948. Merged from trunk r12948.	Alex Rousskov	10 years ago
12582		Prevented STORE_DISK_CLIENT assertions for aborted entries. ... Prevented STORE_DISK_CLIENT assertions for aborted entries. Polished code. To prevent store_client.cc:445: "STORE_DISK_CLIENT == getType()" assertions, rewrote the storeClientNoMoreToSend() function so that it does not send a STORE_MEM_CLIENT to read from disk. Used this opportunity to polish this negative function code and convert it into a positive method. Documented known (and mostly old) StoreEntry::storeClientType() problems.	Alex Rousskov	10 years ago
12581		Better support for things with shared locks that can be open... Better support for things with shared locks that can be opened many times, such as Ipc::StoreMap entries. Maintain a lock counter instead of boolean  opened flag. Better support for things with multipart IDs such as Ipc::StoreMap entries that have an anchor/inode ID and map name. Support searching for shared pages (sh_page).	Alex Rousskov	10 years ago
12580		Re-enabled on-disk collapsing of entries after fixing relate... Re-enabled on-disk collapsing of entries after fixing related code. Since we started writing partial entries, we cannot rely on negative sidNext marking the end of the slice/write sequence. Added a WriteRequest::eof field to signal that end explicitly. Do not leak db slices when write fails or IoState is closed before the write succeeds. Handle store client requesting an offset we have not stored yet. This might happen for collapsed hits (and also if the client is buggy). May need more work to slow the reader down. Do not update various shared stats until the corresponding slot is written.	Alex Rousskov	10 years ago
12579		Relaxed asserts because hits reading collapsed entries may h... Relaxed asserts because hits reading collapsed entries may have zero swap_file_sz.	Alex Rousskov	10 years ago
12578		Improved STORE_MEM_CLIENT detection. Improved STORE_MEM_CLIENT detection. IN_MEMORY mem_status does not guarantee that the entore object is in the memory cache. We may be just loading it from a shared memrory cache, and loading may fail. We may have nibbled at the entry already (although that may not be possible, not sure). The whole memory/disk store_client designation probably needs more work, but the now-removed condition was causing store_client.cc:445: "STORE_DISK_CLIENT == getType()" assertions.	Alex Rousskov	10 years ago
12577		Followup for the StoreEntry::mayStartSwapOut() fix (branch r... Followup for the StoreEntry::mayStartSwapOut() fix (branch r12569). Fixed swapout condition after adding swStarted to allow an already started swapout to continue.	Alex Rousskov	10 years ago
12576		Removing a shared cache entry w/o slices does not count as s... Removing a shared cache entry w/o slices does not count as successful purging because all callers need a new slice, not an available anchor position.	Alex Rousskov	10 years ago
12575		Imprioved sane rock store entry detection to avoid rebuildin... Imprioved sane rock store entry detection to avoid rebuilding coredumps due to out-of-range slot IDs.	Alex Rousskov	10 years ago
12574		Prevent premature swapout aborts when the last local entry r... Prevent premature swapout aborts when the last local entry reader goes away.	Alex Rousskov	10 years ago
12573		Keep anchor.basics.swap_file_sz in sync with slice sizes. Keep anchor.basics.swap_file_sz in sync with slice sizes. The old code updated anchor.basics.swap_file_sz _after_ copying all of the available data into shared memory. An exception in the copying loop (e.g., the map is out of available slots) could prevent that update. For another worker, the entry would then appear to be fully completed (no writer, last slice size stable, and last slice poiner is -1) and that worker would assert due to anchor.basics.swap_file_sz mismatching the sum of slice sizes.	Alex Rousskov	10 years ago
12572		Disconnect StoreEntries before deleting their memory objects... Disconnect StoreEntries before deleting their memory objects. The new cleanup order helps identify the write Rock entry state (reading or writing) and avoid assertions related to state identification bugs (such as unlocking a writing entry for reading). Similar to the memory cache code, we should not disconnect disk entries during shutdown because Store::Root() may be missing by then.	Alex Rousskov	10 years ago

• Older »

Loggerhead is a web-based interface for Breezy
Version: 2.0.1