Hi, the ticket is ready for review. [1] https://github.com/apache/ignite/pull/7728
вт, 28 апр. 2020 г. в 14:39, Sergey Antonov <antonovserge...@gmail.com>: > Maxim, I'm talking about cluster upgrade through cluster stop -> binary > update -> cluster start. > > вт, 28 апр. 2020 г. в 14:37, Maxim Muzafarov <mmu...@apache.org>: > >> Sergey, >> >> Are you talking about a cluster rolling upgrade feature? AFAIK, Apache >> Ignite doesn't support this feature, so why we should care about it? >> >> On Tue, 28 Apr 2020 at 14:32, Sergey Antonov <antonovserge...@gmail.com> >> wrote: >> > >> > Maxim, >> > >> > > should we _reject_ joining nodes which have different >> > From my point of view, it's a breaking change on cluster update. >> > >> > We can get a different inline size in other scenarios too: as I know we >> did >> > some improvements in calculation effective (actual) index inline size. >> > Let's imagine, we have PDS cluster created on the "old" apache ignite >> > version. We decided to upgrade Ignite version and after that, join a new >> > node to the cluster. On a new node, effective inline sizes will be >> > calculated by the optimized algorithm. On the old nodes, the inline size >> > will not be recalculated. It can lead to a difference in inline sizes >> > without IGNITE_MAX_INDEX_PAYLOAD_SIZE option. >> > >> > >> > >> > вт, 28 апр. 2020 г. в 14:22, Maxim Muzafarov <mmu...@apache.org>: >> > >> > > Sergey, Ilya, >> > > >> > > >> > > Since inline size for the `create table` clause not supported yet and >> > > the IGNITE_MAX_INDEX_PAYLOAD_SIZE is the only option, should we >> > > _reject_ joining nodes which have different >> > > IGNITE_MAX_INDEX_PAYLOAD_SIZE value instead for allowing and printing >> > > warning message? Thus we will force users to have the same values of >> > > IGNITE_MAX_INDEX_PAYLOAD_SIZE property. >> > > >> > > >> > > On Tue, 28 Apr 2020 at 14:01, Ilya Kasnacheev < >> ilya.kasnach...@gmail.com> >> > > wrote: >> > > > >> > > > Hello! >> > > > >> > > > Unfortunately and embarrassingly, we still do not support passing >> > > > INLINE_SIZE to CREATE TABLE, at least in 2.8.0. >> > > > This means IGNITE_MAX_INDEX_PAYLOAD_SIZE is the only option to >> create an >> > > > implicit primary key index with specified inline size. >> > > > >> > > > Regards, >> > > > -- >> > > > Ilya Kasnacheev >> > > > >> > > > >> > > > вт, 28 апр. 2020 г. в 02:31, Denis Magda <dma...@apache.org>: >> > > > >> > > > > Hi Sergey, >> > > > > >> > > > > Your changes look useful from the application developer >> perspective. >> > > > > However, I'm curious why would the one change some low-level >> > > > > IGNITE_MAX_INDEX_PAYLOAD_SIZE parameter when it's advised to pass >> > > > > INLINE_SIZE to CREATE TABLE to change the index size cluster-wide. >> > > > > >> > > > > - >> > > > > Denis >> > > > > >> > > > > >> > > > > On Mon, Apr 27, 2020 at 5:38 AM Sergey Antonov < >> > > antonovserge...@gmail.com> >> > > > > wrote: >> > > > > >> > > > > > Hi, Igniters! >> > > > > > >> > > > > > I'd like to share a new small feature in AI [1]. >> > > > > > >> > > > > > For different reasons, the cluster could have a different SQL >> index >> > > > > inline >> > > > > > size [2] on cluster nodes. For example due to >> > > > > > different IGNITE_MAX_INDEX_PAYLOAD_SIZE [3] value on cluster >> nodes. >> > > > > > >> > > > > > The difference in index inline size may lead to performance >> > > degradation. >> > > > > > I think we must compare inline sizes on node join and warn if >> > > difference >> > > > > > found. Also, We should have the ability to check inline sizes on >> > > demand. >> > > > > > >> > > > > > I've implemented this check on node join and new command in >> > > control.sh >> > > > > > >> > > > > > Look at warning message and utility command output: >> > > > > > >> > > > > > Warn message on a node in the cluster during new node join: >> > > > > > >> > > > > > [2020-04-27 15:36:21,185][WARN ][tcp-disco-msg-worker-[6ba0b823 >> > > > > > 127.0.0.1:47502 >> > > > > > >> crd]-#17%cache.CheckIndexesInlineSizeOnNodeJoinMultiJvmTest0%][root] >> > > > > Inline >> > > > > > sizes on local node and node >> 5bf6ca48-34a0-4aff-8db2-c0b6df303d3f are >> > > > > > different. Please drop and create again these indexes to avoid >> > > > > performance >> > > > > > problems with SQL queries. Problem indexes: >> > > > > > >> > > > > > >> > > > > >> > > >> PUBLIC#TEST_TABLE#L_IDX(1,2),PUBLIC#TEST_TABLE#S1_IDX(1,2),PUBLIC#TEST_TABLE#I_IDX(1,2) >> > > > > > >> > > > > > Warn messages on a joining node, if difference found: >> > > > > > [2020-04-27 15:35:17,326][WARN ][tcp-disco-msg-worker-[a86e9cea >> > > > > > 127.0.0.1:47501 >> > > > > > >> ]-#11%cache.CheckIndexesInlineSizeOnNodeJoinMultiJvmTest0%][root] >> > > > > > Inline sizes on local node and node >> > > a86e9cea-63e8-42af-a897-cec4be500001 >> > > > > > are different. Please drop and create again these indexes to >> avoid >> > > > > > performance problems with SQL queries. Problem indexes: >> > > > > > >> > > > > > >> > > > > >> > > >> PUBLIC#TEST_TABLE#L_IDX(1,2),PUBLIC#TEST_TABLE#S1_IDX(1,2),PUBLIC#TEST_TABLE#I_IDX(1,2) >> > > > > > [2020-04-27 15:35:17,326][WARN ][tcp-disco-msg-worker-[a86e9cea >> > > > > > 127.0.0.1:47501 >> > > > > > >> ]-#11%cache.CheckIndexesInlineSizeOnNodeJoinMultiJvmTest0%][root] >> > > > > > Inline sizes on local node and node >> > > a08de16d-df05-48af-a0b9-5596d9c00002 >> > > > > > are different. Please drop and create again these indexes to >> avoid >> > > > > > performance problems with SQL queries. Problem indexes: >> > > > > > >> > > > > > >> > > > > >> > > >> PUBLIC#TEST_TABLE#L_IDX(1,3),PUBLIC#TEST_TABLE#S1_IDX(1,3),PUBLIC#TEST_TABLE#I_IDX(1,3) >> > > > > > >> > > > > > >> > > > > > Utility output, if a difference in inline sizes was found: >> > > > > > >> > > > > > Control utility [ver. 2.9.0-SNAPSHOT#20200427-sha1:DEV] >> > > > > > 2020 Copyright(C) Apache Software Foundation >> > > > > > User: santonov >> > > > > > Time: 2020-04-27T15:32:25.759 >> > > > > > Command [CACHE] started >> > > > > > Arguments: --cache check_index_inline_sizes --yes >> > > > > > >> > > > > > >> > > > > >> > > >> -------------------------------------------------------------------------------- >> > > > > > Found 4 secondary indexes. >> > > > > > 3 index(es) have different effective inline size on nodes. It >> can >> > > lead to >> > > > > > performance degradation in SQL queries. >> > > > > > Index(es): >> > > > > > Full index name: PUBLIC#TEST_TABLE#L_IDX nodes: >> > > > > > [ca1d2bae-89d4-4e8d-ae11-6c68f3900000] inline size: 1, nodes: >> > > > > > [8327abd1-df08-4b97-8720-de95e363e745] inline size: 2 >> > > > > > Full index name: PUBLIC#TEST_TABLE#S1_IDX nodes: >> > > > > > [ca1d2bae-89d4-4e8d-ae11-6c68f3900000] inline size: 1, nodes: >> > > > > > [8327abd1-df08-4b97-8720-de95e363e745] inline size: 2 >> > > > > > Full index name: PUBLIC#TEST_TABLE#I_IDX nodes: >> > > > > > [ca1d2bae-89d4-4e8d-ae11-6c68f3900000] inline size: 1, nodes: >> > > > > > [8327abd1-df08-4b97-8720-de95e363e745] inline size: 2 >> > > > > > >> > > > > > Recommendations: >> > > > > > Check that value of property IGNITE_MAX_INDEX_PAYLOAD_SIZE >> are the >> > > same >> > > > > > on all nodes. >> > > > > > Recreate indexes (execute DROP INDEX, CREATE INDEX commands) >> with >> > > > > > different inline size. >> > > > > > Command [CACHE] finished with code: 0 >> > > > > > Control utility has completed execution at: >> 2020-04-27T15:32:28.025 >> > > > > > Execution time: 2266 ms >> > > > > > >> > > > > > Utility output, if all indexes have the same inline size: >> > > > > > >> > > > > > Control utility [ver. 2.9.0-SNAPSHOT#20200427-sha1:DEV] >> > > > > > 2020 Copyright(C) Apache Software Foundation >> > > > > > User: santonov >> > > > > > Time: 2020-04-27T15:30:20.950 >> > > > > > Command [CACHE] started >> > > > > > Arguments: --cache check_index_inline_sizes --yes >> > > > > > >> > > > > > >> > > > > >> > > >> -------------------------------------------------------------------------------- >> > > > > > Found 2 secondary indexes. >> > > > > > All secondary indexes have the same effective inline size on all >> > > cluster >> > > > > > nodes. >> > > > > > Command [CACHE] finished with code: 0 >> > > > > > Control utility has completed execution at: >> 2020-04-27T15:30:23.428 >> > > > > > Execution time: 2478 ms >> > > > > > >> > > > > > Any objections? >> > > > > > >> > > > > > [1] https://issues.apache.org/jira/browse/IGNITE-12942 >> > > > > > [2] https://apacheignite-sql.readme.io/docs/create-index >> > > > > > [3] >> > > > > > >> > > > > > >> > > > > >> > > >> https://ignite.apache.org/releases/latest/javadoc/org/apache/ignite/IgniteSystemProperties.html#IGNITE_MAX_INDEX_PAYLOAD_SIZE >> > > > > > >> > > > > > -- >> > > > > > BR, Sergey Antonov >> > > > > > >> > > > > >> > > >> > >> > >> > -- >> > BR, Sergey Antonov >> > > > -- > BR, Sergey Antonov > -- BR, Sergey Antonov
