diff options
| author | Suren A. Chilingaryan <csa@suren.me> | 2018-03-24 03:05:47 +0100 |
|---|---|---|
| committer | Suren A. Chilingaryan <csa@suren.me> | 2018-03-24 03:05:47 +0100 |
| commit | 5a15f65db3dfb245919bdd534e93bd711db2eb60 (patch) | |
| tree | 431e6cc7476a0c2e135983a937a9d2e6f184fe5e /docs | |
| parent | c163108c0c0c7b7a4f05da411e98ac0f503e31e0 (diff) | |
| download | ands-5a15f65db3dfb245919bdd534e93bd711db2eb60.tar.gz ands-5a15f65db3dfb245919bdd534e93bd711db2eb60.tar.bz2 ands-5a15f65db3dfb245919bdd534e93bd711db2eb60.tar.xz ands-5a15f65db3dfb245919bdd534e93bd711db2eb60.zip | |
Minor tunning
Diffstat (limited to 'docs')
| -rw-r--r-- | docs/databases.txt | 8 | ||||
| -rw-r--r-- | docs/webservices.txt | 16 |
2 files changed, 22 insertions, 2 deletions
diff --git a/docs/databases.txt b/docs/databases.txt index aa58a2e..bc20f83 100644 --- a/docs/databases.txt +++ b/docs/databases.txt @@ -167,6 +167,10 @@ Master/Slave replication data replicated from master by disabling 'log_slave_updates'. Then, if the slave is converted to master it will automatically start logging. - - + - Further improvements with significant increases of main buffers MYSQL_INNODB_BUFFER_POOL_SIZE and + MYSQL_INNODB_LOG_FILE_SIZE + plus disabling FS caching MYSQL_INNODB_FLUSH_METHOD=ODIRECT. At maximum + I got about 12 MB/s on master and 14 MB/s on the slave (using round-robin access to the source MSSQL + databases). Both ROW and MIXED binlogs give more-or-less the same performance and memory footprint, + but it seems the CPU usage is signifanctly higher (500-800%) in MIXED mode. In ROW mode it was about + 200%.
\ No newline at end of file diff --git a/docs/webservices.txt b/docs/webservices.txt new file mode 100644 index 0000000..8fad471 --- /dev/null +++ b/docs/webservices.txt @@ -0,0 +1,16 @@ + - The users are not directly connected to the services running in OpenShift. There is always + load-balancing HAProxy sitting in between. There is several implications: + * The service will get request from HAProxy IP. I.e. IP-based authentication is not possible + anymore. + * If multiple service replicas running, by default HAProxy will distribute request in round-robin + fashion. I.e. request from the user will be served by different replicas. If we have several running + datbases which are not completely in sync, the user may get confusing changing data. This can be fixed + by setting 'haproxy.router.openshift.io/balance' to 'source' in route metadata. Then, the destination + replica will be determined based on the client IP. + * HAProxy has configured a default timeout. If replica does not send data within '30s' the connection + will be terminated. It can be increased with 'haproxy.router.openshift.io/timeout' + * There is a several ways to configure certiciates for HTTPS services defined by type of tls termination + in the route specification. With 'passthrough' the container is expected to handle certificates itself. + In the edge termination mode, the certificates are configured in the route and HAProxy manages secure + communication with clients and provides unencrypted data to the service in the cluster. +
\ No newline at end of file |