<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0, maximum-scale=1.0" /> <base href="https://hibernate.atlassian.net" />
<title>Message Title</title>
</head>
<body class="jira" style="color: #333; font-family: Arial, sans-serif; font-size: 14px; line-height: 1.429">
<table id="background-table" cellpadding="0" cellspacing="0" width="100%" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt; background-color: #f5f5f5; border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt">
<!-- header here -->
<tr>
<td id="header-pattern-container" style="padding: 0px; border-collapse: collapse; padding: 10px 20px">
<table id="header-pattern" cellspacing="0" cellpadding="0" border="0" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt">
<tr>
<td id="header-avatar-image-container" valign="top" style="padding: 0px; border-collapse: collapse; vertical-align: top; width: 32px; padding-right: 8px"> <img id="header-avatar-image" class="image_fix" src="https://secure.gravatar.com/avatar/6db67528c0dfb637ac5d957a22643392?d=mm&s=48" height="32" width="32" border="0" style="border-radius: 3px; vertical-align: top" />
</td>
<td id="header-text-container" valign="middle" style="padding: 0px; border-collapse: collapse; vertical-align: middle; font-family: Arial, sans-serif; font-size: 14px; line-height: 20px; mso-line-height-rule: exactly; mso-text-raise: 1px"> <a class="user-hover" rel="gunnar.morling" id="email_gunnar.morling" href="https://hibernate.atlassian.net/secure/ViewProfile.jspa?name=gunnar.morling" style="color:#6c797f;; color: #3b73af; text-decoration: none">Gunnar Morling</a> <strong>commented</strong> on <a href="https://hibernate.atlassian.net/browse/HSEARCH-1970" style="color: #3b73af; text-decoration: none"><img src="cid:jira-generated-image-static-improvement-af00ff4d-6308-4169-8bc3-faeb7729a1fe" height="16" width="16" border="0" align="absmiddle" alt="Improvement" /> HSEARCH-1970</a>
</td>
</tr>
</table>
</td>
</tr>
<tr>
<td id="email-content-container" style="padding: 0px; border-collapse: collapse; padding: 0 20px">
<table id="email-content-table" cellspacing="0" cellpadding="0" border="0" width="100%" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt; border-spacing: 0; border-collapse: separate">
<tr>
<!-- there needs to be content in the cell for it to render in some clients -->
<td class="email-content-rounded-top mobile-expand" style="padding: 0px; border-collapse: collapse; color: #fff; padding: 0 15px 0 16px; height: 15px; background-color: #fff; border-left: 1px solid #ccc; border-top: 1px solid #ccc; border-right: 1px solid #ccc; border-bottom: 0; border-top-right-radius: 5px; border-top-left-radius: 5px; height: 10px; line-height: 10px; padding: 0 15px 0 16px; mso-line-height-rule: exactly">
</td>
</tr>
<tr>
<td class="email-content-main mobile-expand " style="padding: 0px; border-collapse: collapse; border-left: 1px solid #ccc; border-right: 1px solid #ccc; border-top: 0; border-bottom: 0; padding: 0 15px 0 16px; background-color: #fff">
<table class="page-title-pattern" cellspacing="0" cellpadding="0" border="0" width="100%" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt">
<tr>
<td style="vertical-align: top;; padding: 0px; border-collapse: collapse; padding-right: 5px; font-size: 20px; line-height: 30px; mso-line-height-rule: exactly" class="page-title-pattern-header-container"> <span class="page-title-pattern-header" style="font-family: Arial, sans-serif; padding: 0; font-size: 20px; line-height: 30px; mso-text-raise: 2px; mso-line-height-rule: exactly; vertical-align: middle"> <a href="https://hibernate.atlassian.net/browse/HSEARCH-1970" style="color: #3b73af; text-decoration: none">Re: Untangle relationship of (DirectoryBased)IndexManager and BackendQueueProvider</a> </span>
</td>
</tr>
</table>
</td>
</tr>
<tr>
<td id="text-paragraph-pattern-top" class="email-content-main mobile-expand comment-top-pattern" style="padding: 0px; border-collapse: collapse; border-left: 1px solid #ccc; border-right: 1px solid #ccc; border-top: 0; border-bottom: 0; padding: 0 15px 0 16px; background-color: #fff; border-bottom: none; padding-bottom: 0">
<table class="text-paragraph-pattern" cellspacing="0" cellpadding="0" border="0" width="100%" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt; font-family: Arial, sans-serif; font-size: 14px; line-height: 20px; mso-line-height-rule: exactly; mso-text-raise: 2px">
<tr>
<td class="text-paragraph-pattern-container mobile-resize-text " style="padding: 0px; border-collapse: collapse; padding: 0 0 10px 0">
<p style="margin: 10px 0 0 0"><a href="http://transcripts.jboss.org/channel/irc.freenode.org/%23hibernate-dev/2015/%23hibernate-dev.2015-08-19.log.html#t2015-08-19T12:26:52" class="external-link" rel="nofollow" style="color: #3b73af; text-decoration: none">IRC discussion</a> on Aug 19th:</p>
<blockquote style="margin: 10px 0 0 0; border-left: 1px solid #ccc; color: #707070; margin-left: 19px; padding: 10px 20px">
<p style="margin: 10px 0 0 0">gmorling sannegrinovero: hey, i have a question regarding BackendQueueProcessor 12:26<br /> gmorling why does it only accept DirectoryBasedIndexManager iso. IndexManager? 12:27<br /> gmorling or hardy, maybe you also know the answer to ^? 12:28<br /> sannegrinovero the concept of tthe BQP is tightly coupled (currently) to the DBIM 12:29<br /> sannegrinovero gmorling: the one practical reason for that which comes to mind is the necessity that the master/slave approach has to control the DP locks 12:29<br /> gmorling hm, ok 12:30<br /> sannegrinovero let me find a pointer.. 12:30<br /> sannegrinovero ah, and that horrible business of the "triggerWorkspaceReconfiguration()" 12:30<br /> sannegrinovero :-/ 12:30<br /> sannegrinovero Ok so the other one is getDirectoryModificationLock() 12:31<br /> sannegrinovero which has to be a shared lock with the one from the Workspace 12:31<br /> sannegrinovero gmorling: there are two open threads on those, aiming to hopefully get rid of them: 12:32<br /> hardy sannegrinovero: knows more about this than Ido 12:32<br /> gmorling i guess what’s confusing me is the interaction of IM and BQP 12:32<br /> sannegrinovero a) the snapshot strategy for mater/slave would avoid need of the Workspace lock at all (there's a JIRA to propose that but emmanuel wasn't convinced on that) 12:32<br /> gmorling i.e. a custom IM implementation (such as the ES one) must be aware of the BQP business 12:32<br /> sannegrinovero hum, but your IM implementation isn't an DBIM right ? 12:33<br /> gmorling no, it isn’t 12:33<br /> gmorling but i’d like to be able to combine ES e.g. with the JMS BQP 12:34<br /> sannegrinovero <span class="error">[to get rid of the triggerWorkspaceReconfiguration() we either get back to the infinispan mailing list to begging they get rid of dynamic reconfiguration, or we ban the feature]</span> 12:34<br /> hardy gmorling: I guess you talk specifically about BackendQueueProcessor#initalize, right? 12:35<br /> gmorling hardy: yes, that’s the method which is tied to DBIM 12:35<br /> sannegrinovero ohh that should be much simpler 12:36<br /> hardy on a first thought I would expect this to be jsut an IndexManaer 12:36<br /> hardy IndexManager 12:36<br /> sannegrinovero right 12:36<br /> sannegrinovero JmsBackendQueueProcessor doesn't need it to be an DBIM 12:36<br /> hardy if a concrete impl needs a specific sub type, it should check the passed index manager 12:36<br /> sannegrinovero +1 12:36<br /> gmorling ok, i will play with that 12:36</p>
<ul>
<li>
<ul>
<li>
<ul>
<li>
mp911de <mp911de!~mp911de@h2.paluch.biz> has quit IRC (Quit: Textual IRC Client: www.textualapp.com) 12:37<br /> sannegrinovero gmorling: we can relax that initialize() contract 12:37<br /> gmorling on a more general note i am wondering why - taking the JGroups backend as an example - on the slave side the IM is involved at all (which then needs to forward to the BQP) instead of that “engine” itself talks to BQP, and only on the master side IM would be involved 12:37<br /> hardy neither the JGroups processor 12:37<br /> hardy (needs a DirectoryBasedIndexManager) 12:38<br /> gmorling k 12:38<br /> hardy so +1 for changing the SPI in this case 12:38<br /> hardy gmorling: do you have a pointer into the code 12:40<br /> gmorling hardy: basically DBIM#performOperations() 12:40<br /> gmorling so here it’s DBIM (or generally, any IM implmenetation) which needs to forward to the BQP 12:41<br /> gmorling i am wondering whether that could not be turned around 12:41<br /> sannegrinovero turned around? 12:42<br /> gmorling i.e. “engine” invokes BQP which then either does its serialization buisness (jgroups slave, jms client) or triggers the IM to do the actual indexing (lucene, jgroups master, jms master) 12:42<br /> hardy but does this not break encapsulation? 12:42<br /> hardy have no real answer atm 12:43<br /> gmorling ok <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/smile.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:43<br /> hardy just thinking out loud 12:43<br /> sannegrinovero me neither, sound like something which needsa whiteboard in Barcelona <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/smile.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:43<br /> gmorling <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/sad.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:43<br /> gmorling where i won’t be unfortunatel 12:43<br /> gmorling y 12:43<br /> hardy i mean in your case the engine would need to know what type/form of setup we are having, right? 12:43<br /> sannegrinovero ho damn 12:44<br /> gmorling hardy: yes, it would; but i think that’d be better than today where this knowledge is spread about IM impls 12:44<br /> sannegrinovero keep in mind that backends like the NRT one require a very high degree of cohesion, for example the IndexWriter is also the source for Readers 12:44<br /> gmorling (which so far was no problem because there were only DBIM impls) 12:44<br /> hardy whereas <b>I think</b> the idea is that the engine just deals with a index manager and the manager knows what to do depemding on what type of index it manages 12:45<br /> hardy again, not sure. Just tjinking out load 12:45<br /> sannegrinovero so one goal is to be able to either use an ultra-efficient NRT-IM or an ElasticSearch-IM without the engine to need knowing too many specifics 12:45<br /> gmorling so to me it’s two (orthogonal) issues here 12:45<br /> gmorling the business of indexing stuff within a specific backend (e.g. lucene, ES, solr) 12:45<br /> gmorling and the business of doing this either locally or sending it to some central master 12:46<br /> gmorling today both aspects seem kind of mangled 12:46<br /> hardy true true 12:46<br /> sannegrinovero well but the way you do "remote" depends onthe kind of backend right? they each have a favourite way 12:46<br /> sannegrinovero Also remember it's 3 parts: 1)indexing 2)querying 3)index maintenance stuff 12:47<br /> gmorling so when writing a new ES IM, i would ideally just need to think about how to do indexing in ES, whereas I’d get the JMS support for “free” in an ideal world 12:47<br /> gmorling sannegrinovero: not sure what you mean by “favorite way”? 12:48<br /> hardy JMS support? 12:48<br /> gmorling e.g. i dont think it makes a difference for jms whether in the end lucene or ES is used 12:48<br /> hardy so you would envison a ES index which gets indexed remotely 12:48<br /> sannegrinovero gmorling: well for example with ES the "default" would be to use the ES REST protocol, not JMS 12:48<br /> gmorling hardy: yes, exactly 12:48<br /> hardy by the engine sending update work to the queue which gets picked up by a master and then applied to ES 12:49<br /> gmorling hardy: right 12:49<br /> sannegrinovero <span class="error">[that was my answer to explain "favourite way"]</span> 12:49<br /> gmorling e.g. to not loose updates 12:49<br /> gmorling by using a persistent queue 12:49<br /> hardy so a complete seperation between any form of serlialization from the type of index 12:49<br /> gmorling sannegrinovero: yes, it’d be the default, but its not inherent to ES that index changes are applied synchronously 12:49<br /> hardy and a free mix and match of serializartion with index “types” 12:50<br /> gmorling instead one might want to have a jms queue in between as lined out above 12:50<br /> gmorling yes, that’s what i’d envision <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/smile.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:50<br /> hardy I guess the data we actually seralize is acgnostic to the index type 12:51<br /> sannegrinovero gmorling: yes I understand the alternative. I only meant to explain that <b>by default</b> the ES IM would have a different style of remoting than the other ones. 12:51<br /> gmorling hardy: yes, it is 12:51<br /> hardy it is just LuceneWork (or Lucene specific data) 12:51<br /> gmorling (afaiu) 12:51<br /> gmorling well, ok, we use Document in there 12:51<br /> gmorling but “semantically” it is independent 12:51<br /> hardy sure, but that is still independent 12:51<br /> hardy even though a different data structure would be benefitial here 12:52<br /> sannegrinovero hardy: that's actually the big question. I wonder if ES will treat our data differently - i.e. say it doesn't take a per-field Analyzer as that'not an option on ES. 12:52<br /> hardy but that is a different issue again 12:52<br /> gmorling right, the usage of Document is more an “ugliness” for me atm. 12:52<br /> gmorling it’s not ideal 12:52<br /> hardy right 12:52<br /> gmorling but i don’t feel its a blocker 12:52<br /> hardy right 12:52<br /> sannegrinovero BTW I noticed we don't need to write "Document" anymore.. Lucene will also accept an "Iterator<Field>" <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/smile.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:52<br /> gmorling Field <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/wink.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:53<br /> sannegrinovero so we can already customize the "value holder" 12:53<br /> sannegrinovero and have it implement the interface .. 12:53<br /> gmorling sannegrinovero: wrt “that <b>by default</b> the ES IM would have a different style of remoting than the other ones.” 12:53<br /> hardy well, in the end you really would like to get away from any of the Lucene classes, but again, the exact design of the (serialized) data is a seperate issue 12:54<br /> gmorling in my picture, the ES IM would not deal with remoting at all 12:54<br /> gmorling it would just call the ES REST api when told to do so 12:54<br /> sannegrinovero hardy: +1 12:54<br /> gmorling whether that is triggered sync or remotely via jms should be decided beforehand somewhere else 12:54
</li>
<li>
guilhermeblanco <guilhermeblanco!~guilherme@204.138.54.2> has joined #hibernate-dev 12:55<br /> sannegrinovero gmorling: ok so you'd have an "ES Backend" and an "ES IndexManager" (which by default uses the ES backend) as two independent components? 12:55<br /> hardy ? 12:55<br /> hardy I guess there there are different types of remoting here 12:55<br /> hardy one which is more in our control which is the support of slave/master setups 12:56<br /> sannegrinovero and then you'd have the option to swap in a JMS backend in the "ES IndexManager", but still use that IM to route Query and index maintenance stuff? 12:56<br /> gmorling sannegrinovero: no, i guess i’d more have a “sync”, “jms” and “jgroups” backend, and a “directory-based” and a “ES-based” IM 12:56<br /> hardy and then with ES there is always another “remote” component when it comes to the actual communication with the ES server 12:56<br /> sannegrinovero gmorling: FYI all of them today support "sync" or "async", that should be orthogonal 12:56<br /> hardy well, we have some reform needs there as well 12:57<br /> gmorling yes, maybe better “local” than “sync” 12:57<br /> hardy at least from a design/description point of view 12:57<br /> sannegrinovero ok but local~directory-based right? 12:57<br /> hardy the terminology we use here is imo very ambigious 12:57<br /> gmorling sannegrinovero: no <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/smile.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 12:58<br /> gmorling local would just invoke the configured IM 12:58<br /> hardy so the options are local, jms, jgroups 12:58<br /> gmorling whichever it is 12:58<br /> sannegrinovero hardy: you mean that sync/async options should have nothing to do with "worker" terminology right? 12:58<br /> hardy which is either do all on the local machine or use of form of serialization 12:58<br /> gmorling maybe directory-based (lucene) or ES 12:58<br /> gmorling whereas “jms” and “jgroups” would not invoke the IM locally but rather go through serialization and remoting 12:59<br /> gmorling and then that remote part there would invoke the IM 12:59<br /> sannegrinovero oh ok now I understand the idea 12:59<br /> gmorling which again would be either lucene or ES 12:59<br /> sannegrinovero But .. 12:59<br /> hardy sannegrinovero: right, the whole concept of workers and sync and async on top of index managers, etc 12:59<br /> hardy it gets all very confusing and there is also overlap between these things 12:59<br /> sannegrinovero I'd like to not rule out other forms of replication: like broadcast all changes and each node keeps its own index in sync locally 12:59<br /> sannegrinovero hardy: +1 13:00
</li>
<li>
smarlow <smarlow!~smarlow@redhat/jboss/smarlow> has joined #hibernate-dev 13:00<br /> gmorling sannegrinovero: sure, that should be doable 13:00
</li>
<li>
andrigtmiller <andrigtmiller!~andrigtmi@c-67-190-58-8.hsd1.co.comcast.net> has joined #hibernate-dev 13:00<br /> gmorling you’d then have a backend which triggers the IM locally <b>and</b> does the broadcasting business 13:00<br /> hardy I like the sound of the idea. It might untanlge a few things 13:01<br /> sannegrinovero gmorling: ok.. I don't dislike the idea, just making sure you keep in mind the other flexible things which we can do today. 13:01<br /> gmorling ok, good to have that feedback 13:01<br /> sannegrinovero for example we really want to implement that pattern for index affinity described on the infinispan wiki 13:01<br /> hardy I’d rather be able to do slightly less in favour of a simpler mroe consistent archtecture and usafge 13:01<br /> gmorling so far i am still in the phase of understanding why things are the way they are 13:01<br /> hardy usage 13:01<br /> gmorling and not the way i think they should be <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/wink.gif" height="16" width="16" align="absmiddle" alt="" border="0" /> 13:01<br /> gmorling hardy: +1 for leaving out mostly exotic use cases if it helps to largely simplify things 13:02<br /> hardy what does all the freedom help us if one does not know how to plug it together 13:02<br /> sannegrinovero hardy: yeah, but I'd still like it to be able to do such things, just possibly hiding the complexity entirely behind a chosen IM 13:02<br /> hardy catering for the 0.0001% of use cases <img class="emoticon" src="https://hibernate.atlassian.net/images/icons/emoticons/wink.gif" height="16" width="16" align="absmiddle" alt="" border="0" />? 13:02<br /> hardy if any 13:02
</li>
<li>
whitingjr <whitingjr!~whitingjr@redhat/jboss/whitingjr> has left #hibernate-dev 13:03<br /> sannegrinovero right, no only for those which make sense, but then rather than giving a configuration guide, to have to set a single IM name which encapsulates the whole setup. 13:03<br /> hardy I think you would gain more happy users by making things simpler and potentially ruling out some super user configurations, than having all the flexibilty in the workd, but you need to study rocket sience to put things togehter 13:04
</li>
<li>
DavideD <DavideD!~ddalto@dab-ell1-h-22-10.dab.02.net> has joined #hibernate-dev 13:04<br /> gmorling my thinking as well 13:04<br /> hardy imo with the next major release (aka 6) we should break with some of the old concepts, try to introduce some sort of consistence view onto the system and make it simpler 13:06<br /> sannegrinovero +1 13:06<br /> hardy looking at Workers and DirectoryProviders for example 13:06<br /> gmorling some clean-up is due for sure 13:07<br /> sannegrinovero let's start with refining the plan already 13:07<br /> sannegrinovero as when the time comes for majors we're usually in a big crunch to not stall releases for too long :-/ 13:07<br /> hardy and I don’t think this is something which you do incremental and with deprecations here and there 13:07<br /> hardy to a certain degree I would envison a bit of a bigger burden for users in favour of a deeper re-design 13:08<br /> sannegrinovero WDYM burden? in terms of migration? 13:09
</li>
<li>
whitingjr <whitingjr!~whitingjr@redhat/jboss/whitingjr> has joined #hibernate-dev 13:10<br /> sannegrinovero hardy: did you happen to play with Apache Camel? 13:10
</li>
<li>
Naros <Naros!~Naros@50-203-221-40-static.hfc.comcastbusiness.net> has joined #hibernate-dev 13:11<br /> hardy sorrry I was distracted for a moment 13:14<br /> sannegrinovero hardy: gmorling: for this conversation, would be good you could have a look here: <a href="https://access.redhat.com/documentation/en-US/Red_Hat_JBoss_Data_Grid/6.5/html-single/Infinispan_Query_Guide/index.html" class="external-link" rel="nofollow" style="color: #3b73af; text-decoration: none">https://access.redhat.com/documentation/en-US/Red_Hat_JBoss_Data_Grid/6.5/html-single/Infinispan_Query_Guide/index.html</a> 13:14<br /> jbossbot Title: Infinispan Query Guide 13:14<br /> hardy right, in terms of migrsation 13:14<br /> hardy it is a bit like the rewrite of the ORM boostrap 13:14<br /> sannegrinovero search for "indexmanager" 13:14<br /> sannegrinovero for example in Infinispan you only have to set a single property: 13:15<br /> sannegrinovero default.indexmanager=org.infinispan.query.indexmanager.InfinispanIndexManager 13:15<br /> sannegrinovero and that will get you: 13:15<br /> sannegrinovero distributed, real-time shared indexes 13:15<br /> sannegrinovero automatic master/slave election 13:15<br /> sannegrinovero failover of masters 13:15<br /> gmorling sure, that’s nice 13:15<br /> sannegrinovero and remoting/serializations setup 13:15<br /> hardy how does Camel come into play here? 13:15<br /> gmorling but you’d never need another BQP than the “local” one in my model to make use of that, right 13:16
</li>
<li>
fjuma <fjuma!~fjuma@redhat/jboss/fjuma> has joined #hibernate-dev 13:16
</li>
<li>
stalep <stalep!~stalep@redhat/jboss/stalep> has joined #hibernate-dev 13:16<br /> sannegrinovero hardy: I was side-tracking a bit, but I would be curious to hear from you if we should "couple" to Camel rather than JMS as it gives users more flexibility w/o making the setup more complex. 13:16<br /> sannegrinovero gmorling: well you have the NRT alternative, and options for synch/asynch 13:17<br /> sannegrinovero gmorling: oh, and it definitley switches the BQP as well.. dynamically. 13:17<br /> gmorling ? 13:18<br /> sannegrinovero it's all logic which is encapsulated in the IM 13:18
</li>
<li>
brmeyer <brmeyer!~brmeyer@redhat/jboss/brmeyer> has joined #hibernate-dev 13:18<br /> sannegrinovero gmorling: maybe clone infinispan and have a look? the /query Maven module is quite small and self-contained, and it's mostly extensions of the Search SPI 13:19<br /> gmorling k, i’ll take a look
</li>
</ul>
</li>
</ul>
</li>
</ul>
</blockquote>
</td>
</tr>
</table>
</td>
</tr>
<tr>
<td class="email-content-main mobile-expand " style="padding: 0px; border-collapse: collapse; border-left: 1px solid #ccc; border-right: 1px solid #ccc; border-top: 0; border-bottom: 0; padding: 0 15px 0 16px; background-color: #fff">
<table id="actions-pattern" cellspacing="0" cellpadding="0" border="0" width="100%" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt; font-family: Arial, sans-serif; font-size: 14px; line-height: 20px; mso-line-height-rule: exactly; mso-text-raise: 1px">
<tr>
<td id="actions-pattern-container" valign="middle" style="padding: 0px; border-collapse: collapse; padding: 10px 0 10px 24px; vertical-align: middle; padding-left: 0">
<table align="left" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt">
<tr>
<td class="actions-pattern-action-icon-container" style="padding: 0px; border-collapse: collapse; font-family: Arial, sans-serif; font-size: 14px; line-height: 20px; mso-line-height-rule: exactly; mso-text-raise: 0px; vertical-align: middle"> <a href="https://hibernate.atlassian.net/browse/HSEARCH-1970#add-comment" target="_blank" title="Add Comment" style="color: #3b73af; text-decoration: none"> <img class="actions-pattern-action-icon-image" src="cid:jira-generated-image-static-comment-icon-2ec4cd5c-a85b-47ec-8472-31483cfc669d" alt="Add Comment" title="Add Comment" height="16" width="16" border="0" style="vertical-align: middle" /> </a>
</td>
<td class="actions-pattern-action-text-container" style="padding: 0px; border-collapse: collapse; font-family: Arial, sans-serif; font-size: 14px; line-height: 20px; mso-line-height-rule: exactly; mso-text-raise: 4px; padding-left: 5px"> <a href="https://hibernate.atlassian.net/browse/HSEARCH-1970#add-comment" target="_blank" title="Add Comment" style="color: #3b73af; text-decoration: none">Add Comment</a>
</td>
</tr>
</table>
</td>
</tr>
</table>
</td>
</tr>
<!-- there needs to be content in the cell for it to render in some clients -->
<tr>
<td class="email-content-rounded-bottom mobile-expand" style="padding: 0px; border-collapse: collapse; color: #fff; padding: 0 15px 0 16px; height: 5px; line-height: 5px; background-color: #fff; border-top: 0; border-left: 1px solid #ccc; border-bottom: 1px solid #ccc; border-right: 1px solid #ccc; border-bottom-right-radius: 5px; border-bottom-left-radius: 5px; mso-line-height-rule: exactly">
</td>
</tr>
</table>
</td>
</tr>
<tr>
<td id="footer-pattern" style="padding: 0px; border-collapse: collapse; padding: 12px 20px">
<table id="footer-pattern-container" cellspacing="0" cellpadding="0" border="0" style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt">
<tr>
<td id="footer-pattern-text" class="mobile-resize-text" width="100%" style="padding: 0px; border-collapse: collapse; color: #999; font-size: 12px; line-height: 18px; font-family: Arial, sans-serif; mso-line-height-rule: exactly; mso-text-raise: 2px">
This message was sent by Atlassian JIRA <span id="footer-build-information">(v7.0.0-OD-01-193#70101-<span title="4fc7e3f5aa699c4da6a6035271b0646e1871bb5e" data-commit-id="4fc7e3f5aa699c4da6a6035271b0646e1871bb5e}">sha1:4fc7e3f</span>)</span>
</td>
<td id="footer-pattern-logo-desktop-container" valign="top" style="padding: 0px; border-collapse: collapse; padding-left: 20px; vertical-align: top">
<table style="border-collapse: collapse; mso-table-lspace: 0pt; mso-table-rspace: 0pt">
<tr>
<td id="footer-pattern-logo-desktop-padding" style="padding: 0px; border-collapse: collapse; padding-top: 3px"> <img id="footer-pattern-logo-desktop" src="cid:jira-generated-image-static-footer-desktop-logo-010449de-9c0c-47b8-9bc1-d2145d924da6" alt="Atlassian logo" title="Atlassian logo" width="169" height="36" class="image_fix" />
</td>
</tr>
</table>
</td>
</tr>
</table>
</td>
</tr>
</table>
</body>
</html>