GEODE-7357: Updating the documentation for member-timeout (#4224)
GEODE-7357: Updating the documentation for member-timeout
Rewriting the documentation for the member-timeout property. The old
documentation did to accurately describe the health check process or the
amount of time it would take to remove an unresponsive member.
diff --git a/geode-docs/reference/topics/gemfire_properties.html.md.erb b/geode-docs/reference/topics/gemfire_properties.html.md.erb
index e294c58..afbfdea 100644
--- a/geode-docs/reference/topics/gemfire_properties.html.md.erb
+++ b/geode-docs/reference/topics/gemfire_properties.html.md.erb
@@ -432,7 +432,14 @@
</tr>
<tr>
<td>member-timeout</td>
-<td><%=vars.product_name%> uses the <code class="ph codeph">member-timeout</code> server configuration, specified in milliseconds, to detect the abnormal termination of members. The configuration setting is used in two ways: 1) First it is used during the UDP heartbeat detection process. When a member detects that a heartbeat datagram is missing from the member that it is monitoring after the time interval of 2 * the value of <code class="ph codeph">member-timeout</code>, the detecting member attempts to form a TCP/IP stream-socket connection with the monitored member as described in the next case. 2) The property is then used again during the TCP/IP stream-socket connection. If the suspected process does not respond to the <em>are you alive</em> datagram within the time period specified in <code class="ph codeph">member-timeout</code>, the membership coordinator sends out a new membership view that notes the member's failure.
+<td><%=vars.product_name%> uses the <code class="ph codeph">member-timeout</code> server configuration, specified in milliseconds, to detect unresponsive members.
+
+<%=vars.product_name%> members use a combination of UDP heartbeat messages and TCP sockets to monitor the health of other members. If a member is suspected to be unreachable, two active checks of the member are performed. Each of those checks waits a maximum of one <code class="ph codeph">member-timeout</code>. If those health checks receive a connection failure or no response within the timeout, the member is considered dead and is removed from the list of members.
+
+This setting also controls how frequently heartbeat messages are sent. They are sent at a frequency of half the <code class="ph codeph">member-timeout</code>.
+
+The total time it takes to remove a member that is not responding to any network traffic is therefore 2 to 3 times the <code class="ph codeph">member-timeout</code>. This is not the minimum time it takes to remove a member from the list of members in the cluster. If the Java process has crashed, but the operating system can still return a connection failure response to the health checks, the crashed member may be removed from the membership list immediately.
+
<p>Valid values are in the range 1000..600000.</p></td>
<td>S, L</td>
<td>5000</td>
